A critical state intelligent judgment method and system based on transformation of contradictory parties

Abstract

This invention discloses an intelligent method and system for determining critical states based on the transformation of contradictory parties, relating to the fields of intelligent state monitoring and dialectical analysis of complex systems. It utilizes distributed heterogeneous sensing units to collect multidimensional time-series data of contradictory parties within a target system in real time. After preprocessing such as normalization and noise reduction, the data is mapped into a quantified sequence of contradictory strength and a dynamically correlated weight sequence based on a preset opposition mapping rule. A recursive dialectical deduction operator is introduced for iterative computation to extract features of quantitative change accumulation, gradient transfer, and potential difference decay, constructing a multidimensional feature vector, which is then input into a pre-trained deep learning model for state recognition. When a critical transformation interval is identified, the critical trigger threshold, confidence level, and transformation direction label are simultaneously output. The system integrates modules for multi-source data acquisition, preprocessing, contradiction mapping quantification, dialectical deduction calculation, critical state recognition, and early warning.

Description

Technical Field

[0001] This invention relates to the field of intelligent state monitoring and dialectical analysis of complex systems, specifically to an intelligent method and system for determining critical states based on the transformation of contradictory sides. Background Technology

[0002] In the operation of complex systems, there are often contradictory and opposing forces that mutually restrict and transform each other. Examples include the "load-strength" contradiction in mechanical systems, the "current-resistance" contradiction in electrical systems, and the "reaction rate-reaction temperature" contradiction in chemical systems. The evolution of these contradictory forces follows the dialectical law of "opposition-unity-transformation." When the quantitative changes of the contradictory forces accumulate to a certain threshold, a qualitative leap will occur, entering a critical transformation state. If this state is not identified and controlled in time, it may lead to system failure, performance degradation, or even paralysis, causing serious economic losses and safety risks.

[0003] In existing technologies, methods for determining the critical state of complex systems can be mainly divided into three categories: The first category is the single threshold method, which sets a fixed parameter threshold and determines the critical state when the system parameter exceeds the threshold. This method does not consider the evolution and interaction of the contradictory parties, resulting in a high misjudgment rate, poor adaptability, and inability to cope with dynamic changes in system parameters. The second category is the linear fitting method, which predicts the parameter change trend and determines the critical state by linearly fitting the time series data of system parameters. This method is only applicable to linearly evolving systems and cannot handle the nonlinear evolution process of the contradictory parties, resulting in low accuracy in identifying critical transformations. The third category is the conventional machine learning method, which identifies critical states by training machine learning models (such as support vector machines, neural networks, etc.). This method only focuses on the statistical characteristics of parameters, does not integrate the dialectical laws of the transformation of the contradictory parties, cannot characterize the direction and potential of contradiction transformation, and has a strong dependence on critical samples and poor generalization ability.

[0004] Furthermore, most existing critical state determination systems employ a centralized data acquisition architecture, which lacks comprehensiveness and synchronicity in data acquisition, failing to accurately capture the synchronous evolution characteristics of contradictory parties. Simultaneously, these systems lack dialectical deduction and dynamic calibration mechanisms, making them ill-suited to the contradictory characteristics of different types of systems, and their determination accuracy significantly declines after long-term operation. In summary, existing technologies suffer from low recognition accuracy, high false positive rates, poor generalization ability, and weak adaptability, failing to meet the practical needs for accurate determination of critical states in complex systems. Therefore, developing a method and system based on the transformation laws of contradictory parties, capable of achieving accurate identification and control of critical states, has significant theoretical and engineering application value. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide an intelligent method and system for determining critical states based on the transformation of contradictory parties. By integrating the dialectical laws of contradiction transformation with intelligent recognition technology, it achieves quantitative analysis of contradiction evolution characteristics, recursive dialectical deduction, and accurate determination of critical states. This solves the problems of low recognition accuracy, high misjudgment rate, poor generalization ability, and weak adaptability in the prior art. At the same time, it realizes hierarchical early warning and dynamic control of critical states, providing a reliable guarantee for the safe and stable operation of complex systems.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method and system for intelligently determining critical states based on the transformation of contradictory parties, comprising the following steps: Step 1: Synchronously collect multi-dimensional physical quantities and state parameters of the target system to construct a time-series data set containing characteristic parameters of the opposing sides. The target system is a complex dynamic system with clearly defined opposing sides, which are two core evolutionary entities that mutually constrain and transform within the system. Characteristic parameters include the operational parameters, structural parameters, energy parameters, and environmental coupling parameters of both sides. The acquisition process uses distributed heterogeneous sensing units for synchronous sampling. The sampling frequency is set to 10Hz-1000Hz according to the evolution rate of the target system to ensure the capture of instantaneous changes in the evolution of the opposing sides. The acquisition duration is no less than 72 hours, forming a complete time-series data set covering the stable, transitional, and critical transformation stages of the contradiction. The format of the time-series data set is [timestamp, parameter 1 of opposing side A, parameter 2 of opposing side A, ..., parameter 1 of opposing side B, parameter 2 of opposing side B, ...], where the timestamp accuracy reaches the millisecond level to ensure the time synchronization of multi-channel data.

[0007] Step 2: Perform normalization and noise reduction preprocessing on the time-series data set to remove abnormal disturbance data and retain the true trend information of the evolution of the contradictory sides; wherein, the normalization process adopts an adaptive normalization algorithm based on the extreme values ​​of the parameters of the contradictory sides, mapping all parameters to the [0,1] interval to eliminate the influence of dimensional differences on subsequent calculations. The normalization formula is:

[0008] In the formula These are the original parameter values. This is the minimum value of the parameter in the time series data set. The maximum value of this parameter in the time series dataset is used for noise reduction. An improved adaptive sliding window filtering algorithm is adopted, and the window size is dynamically adjusted according to the data fluctuation frequency. The adjustment range is 5-25 sampling points. Abnormal disturbance data is removed by calculating the variance threshold of the data within the window. The criterion for judging abnormal data is that it deviates from the mean within the window by more than 3 times the standard deviation. At the same time, an interpolation algorithm based on the correlation characteristics of the parameters of the two opposing sides is used to fill the missing data. The accuracy of filling the missing data is not less than 99.5%, so as to avoid the distortion of the evolution characteristics of the two opposing sides due to missing data.

[0009] Step 3: Based on the preset contradiction-opposition mapping rules, the preprocessed time-series data is mapped into a quantitative contradiction intensity sequence and a dynamic correlation weight sequence for both sides of the contradiction. The preset contradiction-opposition mapping rules are constructed based on the essential attributes of both sides of the contradiction, clarifying the opposition relationship and mapping logic of each set of feature parameters. By establishing an opposition dimension coordinate system for the feature parameters of both sides of the contradiction, each set of time-series data points is mapped to the corresponding coordinate points in the opposition dimension coordinate system. The horizontal axis of the opposition dimension coordinate system represents the normalized value of the parameters of contradiction A, and the vertical axis represents the normalized value of the parameters of contradiction B. The difference between the horizontal and vertical coordinates of the coordinate points represents the instantaneous degree of opposition between the two sides of the contradiction. Based on the distribution of the coordinate points in the opposition dimension coordinate system, the instantaneous opposition intensity value of both sides of the contradiction is calculated using the Euclidean distance formula. These values ​​are then continuously concatenated to form a quantitative opposition intensity sequence. The formula for calculating the instantaneous opposition intensity value is:

[0010] In the formula Let be the intensity of opposition at time t. , Let t be the x and y coordinates of the contradictory party A in the opposing coordinate system. , Let be the x and y coordinates of the contradictory party B in the opposing coordinate system at time t; the dynamic correlation weight is calculated based on the cross-correlation coefficient and lag correlation characteristics of the parameters of the two contradictory parties. The dynamic correlation weight is updated in real time with the evolution sequence, and the update period is consistent with the sampling period. The formula for calculating the dynamic correlation weight is:

[0011] In the formula The dynamic association weights at time t, Let t be the cross-correlation coefficient of the parameters of the two opposing sides at time t, and k be the adjustment coefficient, with a value range of 1.5-3.5. The dynamic correlation weight is used to characterize the degree of mutual constraint and transformation potential of the two opposing sides at different evolution stages. The larger the weight value, the stronger the mutual constraint between the two opposing sides and the greater the transformation potential.

[0012] Step 4: The recursive dialectical deduction operator is used to iteratively calculate the quantified opposition intensity sequence and the dynamic correlation weight sequence, extracting the quantitative accumulation characteristics, gradient transfer characteristics, and potential difference decay characteristics of the two opposing sides in the evolution process. The recursive dialectical deduction operator is a dedicated operator designed based on the law of contradiction transformation. Its core logic is to simulate the dialectical process of "opposition-unity-transformation" between the two opposing sides. Using the current contradiction opposition intensity and correlation weight as input, the recursive dialectical deduction operator completes three steps: opposition intensity correction, correlation weight iteration, and evolution trend prediction. Each iteration retains the deduction result from the previous moment as historical prior information, and combines it with the measured data at the current moment to complete the fusion update. The fusion update formula is:

[0013] In the formula The output value is derived at time t. The output value at time t-1 is the predicted output value, and α is the weighting coefficient of the measured data, with a value range of 0.6-0.8. The iteration termination condition is that the change in the result of N consecutive iterations is less than the preset convergence threshold, where N is a positive integer greater than 3, with a value range of 5-10, and the preset convergence threshold is 0.001-0.005. After the iteration is completed, the cumulative feature of quantitative change is obtained by calculating the cumulative sum of the predicted output sequence, the gradient transfer feature is obtained by calculating the difference between the predicted output values ​​at adjacent times, and the potential difference decay feature is obtained by calculating the decay coefficient of the predicted output sequence. The three features together constitute the core feature set of the contradiction evolution.

[0014] Step 5: Construct a multi-dimensional critical discriminant feature vector based on the quantitative change accumulation feature, gradient transition feature, and potential difference decay feature. Input the multi-dimensional critical discriminant feature vector into the pre-trained contradiction transformation critical identification model for matching operation, and output the contradiction stability interval, contradiction transition interval, or contradiction critical transformation interval currently in which the target system is located. The multi-dimensional critical discriminant feature vector has 9-15 dimensions and, in addition to the three core features mentioned above, also includes derived features such as quantitative change accumulation rate, gradient transition direction, and potential difference decay rate. The feature vector is constructed using a combination of feature normalization and feature fusion to ensure the effectiveness and discriminative power of the feature vector. The pre-trained contradiction transformation critical identification model... This is a dedicated model built on deep residual networks and dialectical feature fusion units. The model structure includes an input layer, a feature extraction layer, a dialectical fusion layer, a classification layer, and an output layer. The input layer receives multi-dimensional critical discriminative feature vectors. The feature extraction layer extracts deep features through residual blocks. The dialectical fusion layer fuses the dialectical features of contradiction evolution. The classification layer uses a softmax activation function to output the probability values ​​of three states. The output layer determines the current state interval of the target system based on the maximum probability. The criteria for determining the state interval are: stable interval probability ≥ 85%, transition interval probability ≥ 60% and stable interval probability < 85%, critical transformation interval probability ≥ 70% and the probabilities of the other two intervals are both < 70%.

[0015] Step 6: When the judgment result is within the critical transformation interval of the contradiction, the critical trigger threshold, critical confidence level, and critical transformation direction label are output simultaneously to complete the intelligent judgment of the critical state based on the transformation of the two sides of the contradiction. Among them, the critical trigger threshold is the minimum opposition intensity value for the two sides of the contradiction to undergo critical transformation, which is dynamically generated by the model according to the current evolution characteristics. The critical confidence level is the reliability of the judgment result, with a value range of 0-100%. The critical transformation direction label is divided into three categories: "transformation dominated by Party A", "transformation dominated by Party B", and "mutual transformation in both directions". The generation of the label is based on the comprehensive judgment of gradient transfer direction and potential difference decay characteristics.

[0016] Furthermore, step 1 involves the synchronous acquisition of multi-dimensional physical quantities and state parameters of the target system, specifically including: Distributed heterogeneous sensing units are used to synchronously collect operational, structural, energy, and environmental coupling parameters of the opposing forces within the target system. Operational parameters include rotational speed, flow rate, pressure, and temperature; structural parameters include deformation, clearance, and stiffness; energy parameters include power, energy consumption, and energy loss; and environmental coupling parameters include ambient temperature, humidity, and vibration amplitude. The number of distributed heterogeneous sensing units is set to 8-32 depending on the scale of the target system, evenly distributed across the key evolution nodes of the opposing forces to ensure the comprehensiveness and representativeness of the collected data. A high-precision time synchronization module is used to achieve time synchronization of all sensing units, with a synchronization error not exceeding 1ms. Multi-channel time-series data streams are formed by aligning the data streams according to a unified timestamp, with the number of channels matching the number of sensing units. The collected multi-channel time-series data is then processed... The data stream is segmented and sliced, with each slice containing synchronous evolution data of the conflicting parties of a preset length. The slice length is set to 1000-10000 sampling points based on the sampling frequency, and a fixed proportion of overlapping data segments is set between adjacent slices to ensure evolution continuity, with an overlap ratio of 20%-40%. Missing value compensation processing is performed on the time-series data after segmentation and slicing. An interpolation algorithm based on the conflict correlation characteristics is used to fill in the missing data. This algorithm calculates the correlation between the parameters of the conflicting parties before and after the missing data, and combines linear interpolation and nonlinear correction to achieve accurate filling of missing data with a filling error of no more than 5%, avoiding distortion of the evolution characteristics of the conflicting parties due to data loss. After the data acquisition is completed, the time-series data set is standardized and uniformly stored in CSV format for easy subsequent preprocessing and computation.

[0017] Furthermore, in step 3, based on a preset contradiction-opposition mapping rule, the preprocessed time-series data is mapped into a quantified sequence of the contradiction intensity and a dynamically correlated weight sequence of the two opposing sides, specifically including: The pre-defined contradiction-opposition mapping rules are formulated by analyzing the essential attributes and evolutionary laws of the contradictory sides in the target system. This clarifies the oppositional relationship of each set of characteristic parameters. For example, in the "load-strength" contradiction of a mechanical system, the load parameter and the strength parameter are opposites; an increase in load corresponds to a decrease in strength, forming a clear oppositional mapping relationship. An oppositional dimension coordinate system is established for the characteristic parameters of the contradictory sides. The dimensions of the coordinate system are consistent with the number of characteristic parameters. For n sets of opposing characteristic parameters, an n-dimensional oppositional dimension coordinate system is constructed. Each set of time-series data points is mapped to the oppositional dimension coordinate system to obtain corresponding coordinate points. Each dimension value of the coordinate point is a normalized value of the corresponding characteristic parameter. Based on the distribution of the coordinate points in the oppositional dimension coordinate system, the instantaneous oppositional strength values ​​of the contradictory sides are calculated using the multi-dimensional Euclidean distance formula. The multi-dimensional Euclidean distance formula is as follows:

[0018] In the formula The feature parameter dimension index. Let be the normalized value of the i-th dimension feature parameter of the contradictory party A at time t. Let be the normalized value of the i-th dimension feature parameter of the opposing side at time t, where n is the number of feature parameter dimensions. The instantaneous opposition intensity values ​​at each time point are continuously concatenated to form a quantified opposition intensity sequence, with the sequence length consistent with the number of sampling points in the time-series data set. Dynamic correlation weights are calculated based on the cross-correlation coefficients and lag correlation characteristics of the parameters of both sides of the contradiction. First, the cross-correlation coefficient matrix of all feature parameters of both sides of the contradiction at time t is calculated, and the maximum cross-correlation coefficient in the matrix is ​​extracted as... Then, the cross-correlation coefficients are mapped to dynamic association weights using the sigmoid function. The formula for calculating the dynamic association weights is:

[0019] In the formula, k is an adjustment coefficient, which is adaptively adjusted according to the contradiction evolution characteristics of the target system. The value range is 1.5-3.5. When the correlation between the two sides of the contradiction is high, r_t approaches 1 and W_t approaches 1. When the correlation between the two sides of the contradiction is low, r_t approaches 0 and W_t approaches 0.5. The dynamic correlation weight is updated in real time with the evolution sequence. The update period is consistent with the sampling period, forming a dynamic correlation weight sequence. The sequence length is consistent with the quantized opposition strength sequence. At the same time, the quantized opposition strength sequence and the dynamic correlation weight sequence are smoothed. The moving average filtering algorithm is used to eliminate high-frequency noise in the sequence. The filtering window size is 3-7 sampling points to ensure the smoothness and continuity of the sequence.

[0020] Furthermore, step 4 employs a recursive dialectical deduction operator to iteratively calculate the quantified opposition strength sequence and the dynamic correlation weight sequence, specifically including: The core structure of the recursive dialectical deduction operator includes an opposition intensity correction unit, a correlation weight iteration unit, and an evolution trend prediction unit. These three units work together to simulate the dialectical evolution process of the two opposing sides of a contradiction: "opposition-unity-transformation." The opposition intensity correction unit is used to correct the quantified opposition intensity value at the current moment, and the correction formula is as follows:

[0021] In the formula This is the corrected value for the intensity of opposition. The opposition strength correction coefficient, with a value ranging from 0.01 to 0.05, is used to correct the opposition strength deviation caused by data noise. The association weight iteration unit is used to iteratively update the dynamic association weight at the current moment. The iteration formula is as follows:

[0022] In the formula The dynamic association weights after iteration. This represents the weight iteration coefficient, with a value range of 0.1-0.3. The cross-correlation coefficient of the parameters of the two opposing sides at time t-1 is used for iterative updates, enabling the dynamic correlation weight to reflect the changes in the correlation between the two opposing sides in real time. The evolution trend prediction unit is used to predict the evolution trend of the contradiction at the next time step based on the corrected opposition intensity value and the iterated dynamic correlation weight. The prediction formula is as follows:

[0023] In the formula This represents the predicted intensity of conflict at time t+1. This is the corrected value of the opposition strength at time t-1. The trend prediction coefficient ranges from 0.2 to 0.4. Each iteration retains the previous time-series results (including the corrected opposition strength value, the iterative dynamic correlation weights, and the predicted value) as historical prior information, and combines this with the current measured data to complete the fusion update. The fusion update formula is as follows:

[0024] In the formula The output value is derived at time t. The output value at time t-1 is the predicted value, where α is the weighting coefficient of the measured data, ranging from 0.6 to 0.8, to ensure a reasonable integration of the measured data and historical prior information. The iteration termination condition is that the change in the result of N consecutive iterations is less than a preset convergence threshold, where N is a positive integer greater than 3, ranging from 5 to 10, and the preset convergence threshold is 0.001 to 0.005. , ... (in When the convergence threshold is reached, the iteration stops; after the iteration is completed, the cumulative characteristic of quantitative change is obtained by calculating the cumulative sum of the deduced output sequence. The cumulative sum formula is:

[0025] In the formula, T represents the total number of iterations; the gradient transfer feature is obtained by calculating the difference between the output values ​​at adjacent time points. The formula for the gradient transfer feature is:

[0026] The potential difference attenuation characteristics are obtained by calculating the attenuation coefficient of the deduced output sequence. The formula for the attenuation coefficient is:

[0027] In the formula To deduce the average value of the output sequence, the three features together constitute the core feature set of the contradiction evolution, providing support for subsequent critical state determination.

[0028] Furthermore, the pre-trained contradiction transformation critical identification model in step 5 is constructed through the following steps: Step 5.1: Collect multiple sets of contradiction evolution sample data from different types of target systems. The sample data covers various complex systems such as mechanical systems, electrical systems, and chemical systems. Each set of sample data contains multi-dimensional time-series data of both sides of the contradiction, as well as corresponding labels for stable state, transition state, and critical transformation state. The total number of sample data sets is no less than 1000 sets, of which stable state samples account for 40%-50%, transition state samples account for 20%-30%, and critical transformation state samples account for 20%-30%, ensuring the diversity and representativeness of the sample data. Perform the same preprocessing, contradiction mapping quantification, and dialectical deduction processing as in Steps 1-4 on the collected sample data, extract the multi-dimensional critical discrimination feature vector of each set of samples, and construct the sample feature set.

[0029] Step 5.2: Divide the sample feature set into a training set, a validation set, and a test set in a ratio of 7:2:1. The training set is used for model parameter training, the validation set is used for model hyperparameter tuning, and the test set is used for model performance verification. The weighted cross-entropy loss function is used to construct the model training objective. The formula for the weighted cross-entropy loss function is:

[0030] In the formula, m is the sample size. Let be the weight of the i-th sample. The weight of the steady-state sample is set to 1.0, the weight of the transitional state sample is set to 1.5, and the weight of the critical transition state sample is set to 2.0 to solve the sample imbalance problem. Let be the true label of the i-th sample (in one-hot encoded form). Let be the predicted probability of the model for the i-th sample.

[0031] Step 5.3: Build the basic model structure based on a deep residual network (ResNet) and a dialectical feature fusion unit. The model structure includes an input layer, a feature extraction layer, a dialectical fusion layer, a classification layer, and an output layer. The dimension of the input layer is consistent with the dimension of the multidimensional critical discriminant feature vector, which is 9-15 dimensions. The feature extraction layer consists of 3-5 residual blocks, each containing 2 convolutional layers, 1 batch normalization layer, and 1 activation function layer. The convolutional kernel size is 3×3, and the activation function uses the ReLU function to extract deep information of the sample features. The dialectical fusion layer is a specially designed feature fusion unit that uses an attention mechanism and a dialectical weight allocation strategy to fuse the core features and derived features of the contradiction evolution, highlighting the weight of the critical transformation features. The fusion formula is:

[0032] In the formula The features are fused, and k is the number of features. The dialectical weight of the j-th feature is obtained through adaptive learning by the model. Let j be the j-th feature; the classification layer uses a fully connected layer and a softmax activation function to output the probability values ​​of the three states; the output layer determines the state label of the sample based on the maximum probability and outputs the decision confidence.

[0033] Step 5.4: During training, introduce prior rules for contradiction transformation to constrain model parameter updates. These prior rules are formulated based on the dialectical laws governing the evolution of the two sides of a contradiction, including rules such as "quantitative changes accumulating to a threshold trigger qualitative changes" and "mutations in related weights predict transformation." By adding these rule constraints, the model training is made to better reflect the actual laws of contradiction transformation. The formula for the rule constraint is:

[0034] By incorporating the rule constraints into the loss function, we obtain the final training loss function:

[0035] In the formula, λ is the weight of the constraint term, and its value ranges from 0.05 to 0.1.

[0036] Step 5.5: Train the model parameters using the Adam optimizer, with a learning rate of 0.001-0.005 and 100-200 iterations. After each training round, verify the model performance using the validation set and adjust the model hyperparameters (including the number of residual blocks, convolutional kernel size, learning rate, etc.). After training, verify the model's decision accuracy and critical recall using the test set. The model performance requirements are: decision accuracy ≥ 98%, critical recall ≥ 99%, and critical misclassification rate ≤ 1%. After meeting the preset indicators, complete the model training and solidify the parameters to obtain the pre-trained contradiction transformation critical identification model.

[0037] Furthermore, when the determination result is within the critical transformation interval of contradiction, the following steps are also included: Step 6.1: Dynamically calibrate the critical trigger threshold. Adjust the threshold value by combining the current environmental disturbance intensity and the system's own attenuation coefficient. The calibration formula is:

[0038] In the formula This is the calibrated critical trigger threshold. The initial output of the model is the critical trigger threshold, where δ is the environmental disturbance influence coefficient, ranging from 0.01 to 0.03, D is the current environmental disturbance intensity (normalized value, ranging from [0,1]), μ is the system attenuation influence coefficient, ranging from 0.02 to 0.04, and λ is the system's own attenuation coefficient. These values ​​are obtained by fitting historical data to ensure the accuracy and adaptability of the critical trigger threshold.

[0039] Step 6.2: Classify the critical warning levels according to the critical confidence level. A confidence level greater than 90% is a Level 1 critical warning, at which point the probability of conflict transformation is extremely high, and control measures need to be taken immediately; 70%-90% is a Level 2 critical warning, at which point conflict transformation is highly likely, and monitoring needs to be strengthened and control measures need to be prepared; 50%-70% is a Level 3 critical warning, at which point conflict transformation is possible, and the trend of conflict evolution needs to be continuously monitored; when the confidence level is below 50%, it is judged as a misjudgment, and data is re-collected for a second judgment. The sampling frequency for the second judgment is increased to 1.5-2 times the original sampling frequency to ensure the reliability of the judgment result.

[0040] Step 6.3: Generate corresponding state control suggestions based on the critical transformation direction label. These suggestions include the adjustment direction, magnitude, and timing of parameters for both conflicting parties, used to suppress or guide the transformation process. When the transformation direction label is "Party A-led Transformation," the control suggestion is to reduce the strength of Party A's parameters and increase the strength of Party B's parameters. The adjustment magnitude is determined based on the opposition strength value and the correlation weight, and the formula for the adjustment magnitude is:

[0041] In the formula, Δx is the adjustment range, and k is the adjustment coefficient, with a value range of 0.05-0.1. When the transformation direction label is "Party B-led transformation", the control suggestion is to reduce the strength of Party B's parameter and increase the strength of Party A's parameter, and the calculation method of the adjustment range is the same as above. When the transformation direction label is "bidirectional mutual transformation", the control suggestion is to adjust the parameters of both parties simultaneously so that the two sides of the contradiction reach a new equilibrium state. The adjustment sequence is determined according to the gradient transfer characteristics, and the parameter of the party with the larger gradient change is adjusted first. After the control suggestion is generated, the adjusted contradiction evolution prediction result is output simultaneously to facilitate the user's evaluation of the control effect.

[0042] Furthermore, the system includes a multi-source data acquisition module, a data preprocessing module, a contradiction mapping quantification module, a dialectical deduction calculation module, a critical state identification module, a critical output and early warning module, a critical calibration and control module, and a model update module. These modules are connected via a high-speed data bus with a data transmission rate of no less than 100Mbps, ensuring real-time performance and stability. The overall system response time does not exceed 500ms, meeting the requirements for real-time determination of critical states in complex systems. The multi-source data acquisition module is used to synchronously acquire multi-dimensional physical quantities and state parameters of the target system, constructing a time-series data set containing characteristic parameters of both contradictory sides. This module adopts a distributed architecture design, which can adapt to target systems of different sizes, supports hot-swapping and expansion of sensing units, and the parameter types acquired can be flexibly configured according to the contradictory characteristics of the target system. The sampling frequency and acquisition duration of the acquired data can be remotely set through the system host computer. After the data is acquired, it is transmitted to the data preprocessing module in real time and stored in the local database. The database adopts a distributed storage architecture with a storage capacity of no less than 1TB, supporting fast data query and retrieval.

[0043] The data preprocessing module is used to perform normalization and noise reduction preprocessing on the time series data set, remove abnormal and disturbing data, and retain the true trend information of the evolution of the two opposing sides. This module includes a normalization processing unit, a noise reduction processing unit, an abnormal data removal unit, and a missing value compensation unit. Each unit works in parallel to improve the preprocessing efficiency. The preprocessed data stream is transmitted to the contradiction mapping and quantification module in real time, and the preprocessing log is stored in the system log database to facilitate subsequent fault investigation and data traceability.

[0044] The contradiction mapping quantification module is used to map preprocessed time series data into a quantified sequence of contradiction intensity and a dynamic correlation weight sequence of the two sides of the contradiction based on preset contradiction opposition mapping rules. This module includes a mapping rule storage unit, a contradiction coordinate system construction unit, a contradiction intensity calculation unit, and a correlation weight calculation unit. The mapping rule storage unit supports users to add, modify, and delete mapping rules to adapt to the contradiction characteristics of different types of target systems. The calculated quantified sequence of contradiction intensity and the dynamic correlation weight sequence are transmitted to the dialectical deduction calculation module in real time.

[0045] The dialectical deduction calculation module is used to perform iterative calculations on the quantified opposition strength sequence and the dynamic correlation weight sequence using a recursive dialectical deduction operator, extracting the quantitative change accumulation features, gradient transfer features, and potential difference decay features of the two opposing sides in the evolution process. This module includes a recursive calculation unit, a historical prior storage unit, a feature extraction unit, and an iterative convergence judgment unit. The historical prior storage unit adopts a high-speed cache design, which can quickly read the deduction results of the previous moment. The iterative convergence judgment unit monitors the iteration process in real time to ensure the stability and convergence of the iterative calculation. The extracted feature set is transmitted to the critical state recognition module in real time.

[0046] The critical state identification module is used to construct a multidimensional critical discrimination feature vector based on the quantitative change accumulation feature, gradient transfer feature, and potential difference decay feature. The multidimensional critical discrimination feature vector is input into a pre-trained contradiction transformation critical identification model for matching operation, and the current state interval of the target system is output. This module includes a feature vector construction unit, a model calling unit, a state determination unit, and a confidence calculation unit. The model calling unit supports the loading and calling of the pre-trained model. The state determination unit determines the state interval based on the probability value output by the model. The confidence calculation unit calculates the reliability of the determination result. The determination result is transmitted to the critical output and early warning module in real time.

[0047] The critical output and early warning module is used to simultaneously output the critical trigger threshold, critical confidence level, critical transformation direction label, and graded early warning information when the judgment result is in the critical transformation interval of the contradiction. This module includes a threshold output unit, a confidence level output unit, a direction label output unit, and an early warning information generation unit. The early warning information supports three methods: audible and visual alarm, SMS alarm, and upper computer pop-up alarm. Users can configure the alarm method and alarm threshold according to their needs. At the same time, this module supports the visualization of the judgment result, showing the contradiction evolution trend and critical state information through line charts, bar charts, etc.

[0048] Furthermore, the multi-source data acquisition module specifically includes a distributed heterogeneous sensing unit, a timestamp alignment unit, a data segmentation and slicing unit, a missing value compensation unit, and a data storage unit; wherein: The distributed heterogeneous sensing unit consists of various types of sensors, including temperature sensors, pressure sensors, displacement sensors, vibration sensors, and power sensors. The accuracy class of the sensors is no less than 0.1, and the measurement range can be flexibly configured according to the target system parameter range. For example, the temperature sensor has a measurement range of -50℃ to 200℃, the pressure sensor has a measurement range of 0 to 10MPa, and the displacement sensor has a measurement range of 0 to 100mm. The sensors adopt an industrial-grade protection design with a protection level of no less than IP67, which can adapt to harsh industrial environments. The sampling frequency of the sensors can be independently set through the system host computer, with a setting range of 10Hz to 1000Hz.

[0049] The timestamp alignment unit uses a high-precision time synchronization module (such as a GPS synchronization module or a Beidou synchronization module) to achieve time synchronization of all distributed heterogeneous sensing units. The time synchronization error does not exceed 1ms. This unit aligns the data collected by each sensor according to a unified timestamp to form a multi-channel time-series data stream. The data stream format is [timestamp, contradictory A parameter 1, contradictory A parameter 2, ..., contradictory B parameter 1, contradictory B parameter 2, ...]. The timestamp accuracy reaches the millisecond level, ensuring the time synchronization of multi-channel data.

[0050] The data segmentation and slicing unit is used to segment time series data and set overlapping segments. This unit supports user-defined slice length and overlap ratio. The slice length can be set from 1,000 to 10,000 sampling points, and the overlap ratio can be set from 20% to 40%. The segmented and sliced ​​time series data is transmitted to the missing value compensation unit in batches and stored in the data storage unit at the same time.

[0051] The missing value compensation unit is used to fill missing data using a contradictory correlation interpolation algorithm. This unit first performs missing value detection on the time series data with a detection accuracy of 100%, which can accurately identify the location and quantity of missing data. Then, by calculating the correlation between contradictory parameters before and after the missing data, and combining linear interpolation and nonlinear correction, the missing data is accurately filled with a filling error of no more than 5%. The filled time series data is then transmitted to the data preprocessing module.

[0052] The data storage unit adopts a distributed storage architecture, consisting of multiple storage nodes with a storage capacity of no less than 1TB. It supports real-time writing and fast reading of data. The stored data includes raw collected data, segmented data, and data with missing values ​​filled in. The data storage formats are CSV and binary. The CSV format is used for data query and analysis, while the binary format is used for fast data storage and transmission. At the same time, the unit supports automatic data backup, with a backup cycle that can be set to 12-24 hours to ensure data security.

[0053] Furthermore, the dialectical deduction calculation module includes a recursive operation unit, a historical prior storage unit, a feature extraction unit, an iterative convergence judgment unit, and a deduction log recording unit; wherein: The recursive operation unit is used to perform iterative calculations of the recursive dialectical deduction operator. This unit adopts FPGA hardware acceleration design, with a calculation speed of no less than 10^6 times / second. It can quickly complete the three-step calculations of opposition strength correction, correlation weight iteration and evolution trend prediction. The formula and steps of the iterative calculation strictly follow the provisions of claim 4 to ensure the accuracy of the deduction results.

[0054] The historical prior storage unit adopts a high-speed cache (SRAM) design with a cache capacity of no less than 16MB. It can quickly read the inference results of the previous moment (including the corrected opposition strength value, the iterative dynamic correlation weight and the predicted value) as historical prior information. The cache read and write speed is no less than 100MB / s to ensure the real-time performance of iterative calculations. At the same time, this unit supports persistent storage of historical prior information. When the system is powered off, the historical prior information is automatically saved to the local storage medium to avoid data loss.

[0055] The feature extraction unit is used to extract cumulative features of quantitative changes, gradient transfer features, and potential decay features. This unit adopts a dedicated feature extraction algorithm, which can quickly calculate the cumulative sum, gradient difference, and decay coefficient of the inferred output sequence. The feature extraction accuracy is no less than 99.8%. After the extracted feature set is organized in a fixed format, it is transmitted to the critical state recognition module in real time and stored in the feature database to facilitate subsequent model updates and performance optimization.

[0056] The iterative convergence judgment unit is used to determine whether the iteration has reached the preset convergence condition and control the termination of the iteration. This unit monitors the change in the result of each iteration in real time and compares the change with the preset convergence threshold. When the change in the result of N consecutive iterations is less than the convergence threshold, an iteration termination signal is sent immediately to control the recursive operation unit to stop iterating. The value of N is in the range of 5-10, and the value of the convergence threshold is in the range of 0.001-0.005. Users can flexibly adjust the size of N and the convergence threshold according to the evolution characteristics of the target system.

[0057] The deduction log recording unit is used to record the entire process of iterative calculation, including the time of each iteration, the opposition strength value, the correlation weight value, the deduction output value and the change amount. The accuracy of the log recording reaches the millisecond level. The deduction log is stored in the system log database in real time, which facilitates subsequent fault diagnosis, deduction process tracing and algorithm optimization. The log data is kept for no less than 3 months.

[0058] Furthermore, it also includes a critical calibration and control module, which is used to dynamically calibrate the critical trigger threshold, classify early warning levels according to the critical confidence level, and generate parameter control suggestions for both sides of the contradiction based on the direction of critical transformation. The critical calibration and control module specifically includes a threshold calibration unit, an early warning level classification unit, and a control suggestion generation unit. The threshold calibration unit adopts an adaptive calibration algorithm, which combines the current environmental disturbance intensity and the system's own attenuation coefficient to correct the critical trigger threshold in real time, with a calibration accuracy of not less than 99%. The early warning level classification unit automatically classifies the critical early warning levels into first-level, second-level, and third-level based on the critical confidence level and generates corresponding early warning information. The control suggestion generation unit generates targeted parameter adjustment schemes based on the critical transformation direction label and the characteristics of contradiction evolution, including adjustment direction, adjustment magnitude, and adjustment sequence. The adjustment schemes can be directly output to the control execution mechanism of the target system to realize automatic control of contradiction transformation.

[0059] The system also includes a model update module, which periodically collects new contradiction evolution data to incrementally update the pre-trained model, optimizing the model's critical judgment accuracy and generalization ability under different scenarios, and ensuring long-term stable operation. Specifically, the model update module includes a data acquisition unit, a data processing unit, a model training unit, and a model replacement unit. The data acquisition unit periodically collects new contradiction evolution data of the target system, with an acquisition cycle of 1-3 months and a data volume of no less than 100 sets. The data processing unit preprocesses the newly collected data, performs contradiction mapping quantification and dialectical deduction processing, and extracts new feature sets. The model training unit uses an incremental training algorithm to update the parameters of the pre-trained model based on the new feature sets. The incremental training iterations are 50-100 rounds, with a learning rate of 0.0005-0.001 to avoid model overfitting. After the new model is trained and passes performance verification, the model replacement unit automatically replaces the old model in the system. The replacement process does not affect the normal operation of the system, ensuring that the system's judgment accuracy remains at a high level. Simultaneously, the model update module supports model version management, retaining historical model versions for quick rollback to the old model when problems arise with the new model.

[0060] This invention provides an intelligent method and system for determining critical states based on the transformation of contradictory parties, which has the following beneficial effects: 1. By innovatively integrating the dialectical law of contradiction transformation with intelligent recognition technology, a critical state determination system based on the transformation of the two sides of a contradiction has been constructed. This system overcomes the shortcomings of existing technologies that only focus on the statistical characteristics of parameters and ignore the law of contradiction evolution, thereby improving the accuracy and reliability of critical state recognition.

[0061] 2. A dedicated recursive dialectical deduction operator was designed, which can simulate the evolutionary process of "opposition-unity-transformation" between the two sides of a contradiction, accurately extract the quantitative accumulation, gradient transfer and potential difference decay characteristics of the contradiction evolution, realize the quantitative characterization of the contradiction evolution law, and provide a brand-new technical path for critical state determination.

[0062] 3. A pre-trained critical identification model for contradiction transformation was constructed, and prior rules for contradiction transformation were introduced to constrain the model training. This solved the problems of poor generalization ability and strong dependence on critical samples in existing machine learning models. It can adapt to different types of complex systems and has strong versatility and adaptability.

[0063] 4. A distributed multi-source data acquisition architecture was designed to achieve synchronous acquisition of multi-dimensional parameters of both sides of the contradiction. Combined with an adaptive preprocessing algorithm, high data quality was ensured, providing reliable support for subsequent calculations. At the same time, the system has a dynamic calibration and model incremental update mechanism, which can adapt to the dynamic changes of the system and ensure the judgment accuracy in long-term operation.

[0064] 5. It realizes graded early warning and dynamic control of critical state, and can divide the early warning level according to the critical confidence level and generate targeted control suggestions according to the transformation direction. It can not only identify the critical state in time, but also actively guide or suppress the transformation of contradictions, effectively avoid system failure, and improve the safety, stability and service life of the system. Attached Figure Description

[0065] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0066] Figure 1 This is a general flowchart of the overall method of the present invention; Figure 2 This is a detailed flowchart of the data acquisition process for this invention; Figure 3 This is a flowchart illustrating the contradiction mapping and quantification process of the present invention. Figure 4 This is a flowchart of the recursive dialectical deduction process of the present invention; Figure 5 This is a flowchart of the criticality determination and early warning control process of the present invention. Detailed Implementation

[0067] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses consistent with some aspects of this disclosure as detailed in the appended claims.

[0068] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0069] Example 1: Implementation of an intelligent method for determining critical states based on the transformation of contradictory parties

[0070] This embodiment takes the gear transmission system of a heavy machinery as the target system. The core contradiction of this system is "gear load (contradiction A) - gear strength (contradiction B)". When the cumulative change in gear load exceeds the bearing limit of gear strength, critical failures such as gear fracture will occur. The specific implementation steps of the method of this invention are as follows: Step 1: Simultaneously collect time-series data from both sides of the conflict from multiple dimensions. Sixteen distributed heterogeneous sensing units are employed, evenly distributed at key nodes of the gear transmission system. These include: four temperature sensors for acquiring gear surface temperature (an auxiliary parameter); four pressure sensors for acquiring gear meshing load (a core parameter); four displacement sensors for acquiring gear deformation (an auxiliary parameter); and four hardness sensors for acquiring gear tooth surface hardness (a core parameter). The sensors have an accuracy class of 0.1. The temperature sensor measurement range is -50℃ to 200℃, the pressure sensor measurement range is 0-10MPa, the displacement sensor measurement range is 0-100mm, and the hardness sensor measurement range is 100-1000HV.

[0071] The Beidou synchronization module is used to achieve time synchronization of all sensing units. The time synchronization error does not exceed 1ms. The sampling frequency is set to 100Hz and the collection time is 72 hours. A total of 72×3600×100=25,920,000 sampling points are collected to form a time series data set, in the format of [timestamp (ms), gear load (MPa), gear surface temperature (°C), gear deformation (mm), gear tooth surface hardness (HV)].

[0072] The acquired multi-channel time-series data stream was segmented and sliced, with a slice length of 5000 sampling points and an overlap ratio of 30% between adjacent slices, resulting in a total of 5184 slices. Missing values ​​were detected in the segmented time-series data, and the amount of missing data was found to be 0.3%. An interpolation algorithm based on contradictory correlation characteristics was used to fill the missing data, and the filling error was 3.2%, which was lower than the preset threshold of 5%.

[0073] Step 2: Normalization, noise reduction, and missing data compensation preprocessing

[0074] An adaptive normalization algorithm based on the extreme values ​​of the parameters of the contradictory parties is adopted to map all parameters to the interval [0,1]. The normalization formula is as follows: Among them, the minimum gear load is 1MPa and the maximum is 9MPa; the minimum gear surface temperature is 25℃ and the maximum is 150℃; the minimum gear deformation is 0.1mm and the maximum is 5mm; the minimum gear tooth surface hardness is 300HV and the maximum is 800HV.

[0075] An improved adaptive sliding window filtering algorithm was used for noise reduction. The window size was dynamically adjusted according to the data fluctuation frequency, ranging from 5 to 25 sampling points. Abnormal data disturbances were removed by calculating the variance threshold of the data within the window (set to 0.01), and a total of 1296 abnormal data were removed, accounting for 0.005% of the total data. An interpolation algorithm based on contradictory correlation characteristics was used to fill in the missing data, and the data integrity reached 99.7% after filling.

[0076] Step 3: Quantification of Contradictory Mapping

[0077] Based on the contradictory relationship between "gear load and gear strength", a contradictory mapping rule is established: as the gear load increases, the gear strength (hardness and resistance to deformation) decreases, and the two are negatively correlated. A two-dimensional coordinate system is established, with the horizontal axis representing the normalized value of the gear load and the vertical axis representing the comprehensive normalized value of the gear strength (calculated by weighting gear deformation and tooth surface hardness, with weights of 0.4 and 0.6, respectively).

[0078] Based on the distribution of coordinate points in the opposing coordinate system, and using the two-dimensional Euclidean distance formula, the instantaneous intensity of the conflict between the two sides is calculated. The formula is as follows: The instantaneous opposition intensity values ​​at each moment are continuously concatenated to form a quantized opposition intensity sequence with a sequence length of 25,920,000 data points.

[0079] The dynamic correlation weight is calculated based on the cross-correlation coefficient and hysteresis characteristics between gear load and gear strength parameters. The calculated cross-correlation coefficient r_t ranges from 0.6 to 0.9, and the adjustment coefficient k is set to 2.5. The formula for calculating the dynamic correlation weight is as follows: The obtained dynamic correlation weight sequence ranges from 0.81 to 0.95, indicating that the correlation between gear load and gear strength is high and the potential for transformation is large.

[0080] The quantitative opposition strength sequence and the dynamic correlation weight sequence are smoothed by using a moving average filtering algorithm with a filter window size of 5 sampling points. The noise of the smoothed sequence is significantly reduced, and it can clearly reflect the evolution trend of the two opposing sides.

[0081] Step 4: Recursive Dialectical Deduction

[0082] Iterative calculations were performed using a recursive dialectical deduction operator. The opposition intensity correction coefficient ΔI was set to 0.03, the weight iteration coefficient β was set to 0.2, the trend prediction coefficient γ was set to 0.3, the measured data weight coefficient α was set to 0.7, and the iteration termination condition was that the change in the result of 8 consecutive iterations was less than the preset convergence threshold of 0.003.

[0083] During the iterative calculation, each iteration retains the deduction result from the previous moment as historical prior information, and combines it with the measured data at the current moment to complete the fusion update. The fusion update formula is as follows: After iterative calculations, when the iteration reaches the 8th iteration, the change in the result of the 8 consecutive iterations is less than 0.003, which satisfies the convergence condition, and the iteration stops.

[0084] After iteration, the core features of the contradiction evolution were extracted: the quantitative change accumulation feature S=129600.8, the gradient transfer feature G_t ranges from 0.001 to 0.008, and the potential difference decay feature λ=0.0025. This indicates that the quantitative change of gear load is continuously accumulating, the gradient transfer direction is that the load increases and the intensity decreases, the potential difference decays slowly, and the contradiction is gradually evolving towards the critical transformation direction.

[0085] Step 5: Critical State Identification

[0086] Based on the extracted quantitative change accumulation features, gradient transfer features, and potential difference decay features, a 12-dimensional critical discrimination feature vector is constructed, including derived features such as quantitative change accumulation value, quantitative change accumulation rate, gradient transfer value, gradient transfer direction, potential difference decay coefficient, and potential difference decay rate.

[0087] The multidimensional critical discriminant feature vector is input into the pre-trained critical identification model of contradiction transformation. This model is built on ResNet-18 deep residual network and dialectical feature fusion unit. Prior rules of contradiction transformation are introduced during training. The model has a judgment accuracy of 98.5%, a critical recall rate of 99.2%, and a critical misclassification rate of 0.8%.

[0088] The model outputs the following state interval probabilities: stable interval probability 12%, transition interval probability 18%, and critical transformation interval probability 70%. Based on the judgment criteria, the target system is determined to be in the critical transformation interval of the contradiction, with a judgment confidence level of 85%.

[0089] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for intelligently determining critical states based on the transformation of contradictory parties, characterized in that, Includes the following steps: Step 1: Synchronously collect multi-dimensional physical quantities and state parameters of the target system to construct a time-series data set containing characteristic parameters of both contradictory sides; Step 2: Perform normalization and noise reduction preprocessing on the time series data set to remove abnormal disturbance data and retain the true trend information of the evolution of the two contradictory sides; Step 3: Based on the preset contradiction and opposition mapping rules, the preprocessed time series data is mapped into a quantitative contradiction intensity sequence and a dynamic correlation weight sequence of the two sides of the contradiction; Step 4: Use the recursive dialectical deduction operator to iteratively calculate the quantified opposition intensity sequence and the dynamic correlation weight sequence to extract the quantitative change accumulation characteristics, gradient transfer characteristics and potential difference decay characteristics of the two sides in the evolution process. Step 5: Construct a multidimensional critical discrimination feature vector based on the quantitative change accumulation feature, gradient transfer feature and potential difference decay feature. Input the multidimensional critical discrimination feature vector into the pre-trained contradiction transformation critical identification model for matching operation, and output the contradiction stability interval, contradiction transition interval or contradiction critical transformation interval of the target system at present. Step 6: When the judgment result is the critical transformation interval of the contradiction, the critical trigger threshold, critical confidence and critical transformation direction label are output simultaneously to complete the intelligent judgment of the critical state based on the transformation of the two sides of the contradiction.

2. The method according to claim 1, characterized in that, Step 1 involves the synchronous acquisition of multi-dimensional physical quantities and state parameters of the target system, specifically including: Distributed heterogeneous sensing units are used to synchronously collect the operating parameters, structural parameters, energy parameters, and environmental coupling parameters of the opposing forces within the target system. A high-precision time synchronization module is used to achieve time synchronization of all sensing units, and a multi-channel time-series data stream is formed by aligning according to a unified timestamp. The number of channels in the data stream is consistent with the number of sensing units. The acquired multi-channel time-series data stream is segmented and sliced, with each slice containing a preset length of synchronous evolution data of the conflicting parties. Missing value compensation is performed on the segmented and sliced ​​time series data, and an interpolation algorithm based on contradictory correlation characteristics is used to fill the missing data; The time series data set is standardized and stored in CSV format.

3. The method according to claim 1, characterized in that, Step 3, based on a preset contradiction-opposition mapping rule, maps the preprocessed time-series data into a quantified sequence of the opposing forces and a dynamically correlated weight sequence, specifically including: The pre-defined contradiction-opposition mapping rules are formulated by analyzing the essential attributes and evolutionary laws of the two contradictory sides of the target system, and clarify the opposition relationship of each set of characteristic parameters; Establish a coordinate system of opposing dimensions for the characteristic parameters of the two opposing sides. The dimensions of the coordinate system are consistent with the number of characteristic parameters. For n sets of opposing characteristic parameters, construct an n-dimensional coordinate system of opposing dimensions. Map each set of time series data points to the coordinate system of opposing dimensions to obtain the corresponding coordinate points. Each dimension value of the coordinate point is a normalized value of the corresponding characteristic parameter. Based on the distribution of coordinate points in the opposing dimension coordinate system, and using the multidimensional Euclidean distance formula, the instantaneous intensity of the conflict between the two sides is calculated. The multidimensional Euclidean distance formula is as follows: In the formula The feature parameter dimension index. Let be the normalized value of the i-th dimension feature parameter of the contradictory party A at time t. Let t be the normalized value of the i-th dimension feature parameter of the contradictory party B at time t, and n be the number of dimensions of the feature parameter. The instantaneous opposition intensity values ​​of each time point are continuously concatenated to form a quantified opposition intensity sequence. The sequence length is consistent with the number of sampling points in the time series data set. The dynamic correlation weight is calculated based on the cross-correlation coefficients and lag correlation characteristics of the parameters of the two contradictory parties. First, the cross-correlation coefficient matrix of all characteristic parameters of the two contradictory parties at time t is calculated, and the maximum cross-correlation coefficient in the matrix is ​​extracted as the weight. Then, the cross-correlation coefficients are mapped to dynamic association weights using the sigmoid function. The formula for calculating the dynamic association weights is: In the formula, k is the adjustment coefficient, which is adaptively adjusted according to the contradiction evolution characteristics of the target system, and the value range is 1.5-3.5; the dynamic correlation weight is updated in real time with the evolution time sequence, and the update period is consistent with the sampling period, forming a dynamic correlation weight sequence, and the sequence length is consistent with the quantification of the opposition intensity sequence.

4. The method according to claim 1, characterized in that, Step 4 employs a recursive dialectical deduction operator to iteratively calculate the quantified opposition strength sequence and the dynamic correlation weight sequence, specifically including: The core structure of the recursive dialectical deduction operator includes an opposition intensity correction unit, a correlation weight iteration unit, and an evolution trend prediction unit. These three units work together to simulate the dialectical evolution process of the two opposing sides. The opposition intensity correction unit is used to correct the quantified opposition intensity value at the current moment, and the correction formula is: In the formula This is the corrected value for the intensity of opposition. The opposition strength correction coefficient, with a value ranging from 0.01 to 0.05, is used to correct the opposition strength deviation caused by data noise. The association weight iteration unit is used to iteratively update the dynamic association weight at the current moment. The iteration formula is as follows: In the formula The dynamic association weights after iteration. This represents the weight iteration coefficient, with a value range of 0.1-0.

3. The cross-correlation coefficient of the parameters of the two opposing sides at time t-1 is used for iterative updates, enabling the dynamic correlation weight to reflect the changes in the correlation between the two opposing sides in real time. The evolution trend prediction unit is used to predict the evolution trend of the contradiction at the next time step based on the corrected opposition intensity value and the iterated dynamic correlation weight. The prediction formula is as follows: In the formula This represents the predicted intensity of conflict at time t+1. This is the corrected value of the opposition strength at time t-1. The trend prediction coefficient ranges from 0.2 to 0.

4. Each iteration retains the previous time-series results (including the corrected opposition strength value, the iterative dynamic correlation weights, and the predicted value) as historical prior information, and combines this with the current measured data to complete the fusion update. The fusion update formula is as follows: In the formula The output value is derived at time t. The output value at time t-1 is the predicted value, α is the weighting coefficient of the measured data, and its value ranges from 0.6 to 0.8; the iteration termination condition is that the change in the result of N consecutive iterations is less than the preset convergence threshold, N is a positive integer greater than 3, and its value ranges from 5 to 10, and the preset convergence threshold is 0.001 to 0.005; after the iteration is completed, the cumulative characteristic of quantitative change is obtained by calculating the cumulative sum of the predicted output sequence. The cumulative sum formula is: In the formula, T represents the total number of iterations; the gradient transfer feature is obtained by calculating the difference between the output values ​​at adjacent time points. The formula for the gradient transfer feature is: The potential difference attenuation characteristics are obtained by calculating the attenuation coefficient of the derived output sequence. The formula for the attenuation coefficient is: In the formula To deduce the average value of the output sequence, the three features together constitute the core feature set of the contradiction evolution, providing support for subsequent critical state determination.

5. The method according to claim 1, characterized in that, The pre-trained contradiction transformation critical identification model in step 5 is constructed through the following steps: Step 5.1: Collect multiple sets of contradiction evolution sample data from different types of target systems. The sample data covers various complex systems such as mechanical systems, electrical systems, and chemical systems. Each set of sample data contains multi-dimensional time-series data of both sides of the contradiction, as well as corresponding labels for stable state, transition state, and critical transformation state. Perform the same preprocessing, contradiction mapping quantification, and dialectical deduction processing as in Steps 1-4 on the collected sample data. Extract the multi-dimensional critical discrimination feature vector of each set of samples to construct a sample feature set. Step 5.2: Divide the sample feature set into a training set, a validation set, and a test set in a ratio of 7:2:

1. The training set is used for model parameter training, the validation set is used for model hyperparameter tuning, and the test set is used for model performance verification. The weighted cross-entropy loss function is used to construct the model training objective. The formula for the weighted cross-entropy loss function is: In the formula, m is the sample size. Let be the weight of the i-th sample. The weight of the steady-state sample is set to 1.0, the weight of the transitional state sample is set to 1.5, and the weight of the critical transition state sample is set to 2.0 to solve the sample imbalance problem. Let i be the true label of the i-th sample. Let be the predicted probability of the model for the i-th sample. Step 5.3: Build the basic model structure based on deep residual networks and dialectical feature fusion units. The model structure includes an input layer, a feature extraction layer, a dialectical fusion layer, a classification layer, and an output layer. The dimension of the input layer is consistent with the dimension of the multidimensional critical discriminant feature vector, which is 9-15 dimensions. The feature extraction layer consists of 3-5 residual blocks. Each residual block contains 2 convolutional layers, 1 batch normalization layer, and 1 activation function layer. The convolutional kernel size is 3×3, and the activation function is the ReLU function, which is used to extract deep information of sample features. The dialectical fusion layer is a specially designed feature fusion unit that employs an attention mechanism and a dialectical weight allocation strategy to fuse the core and derived features of contradiction evolution, highlighting the weight of critical transformation features. The fusion formula is as follows: In the formula The features are fused, and k is the number of features. The dialectical weight of the j-th feature is obtained through adaptive learning by the model. For the j-th feature; The classification layer uses a fully connected layer and a softmax activation function to output the probability values ​​of the three states; the output layer determines the state label of the sample based on the maximum probability and outputs the decision confidence level. Step 5.4: During training, introduce prior rules for contradiction transformation to constrain model parameter updates. These prior rules are formulated based on the dialectical laws governing the evolution of contradictory parties. By adding rule constraints, the model training is made to better reflect the actual laws of contradiction transformation. The formula for the rule constraints is: By incorporating the rule constraints into the loss function, we obtain the final training loss function: In the formula, λ is the weight of the constraint term, and its value ranges from 0.05 to 0.

1. Step 5.5: Train the model parameters using the Adam optimizer, with a learning rate of 0.001-0.005 and 100-200 iterations. After each training round, verify the model performance using the validation set and adjust the model hyperparameters. After training, verify the model's decision accuracy and critical recall using the test set. The model performance requirements are: decision accuracy ≥ 98%, critical recall ≥ 99%, and critical misclassification rate ≤ 1%. After meeting the preset indicators, complete the model training and solidify the parameters to obtain the pre-trained contradiction transformation critical identification model.

6. The method according to claim 1, characterized in that, When the determination result is within the critical transformation interval of contradiction, the following steps are also included: Step 6.1: Dynamically calibrate the critical trigger threshold. Adjust the threshold value by combining the current environmental disturbance intensity and the system's own attenuation coefficient. The calibration formula is: In the formula This is the calibrated critical trigger threshold. δ is the critical trigger threshold of the initial output of the model, δ is the environmental disturbance influence coefficient, with a value range of 0.01-0.03, D is the current environmental disturbance intensity, μ is the system attenuation influence coefficient, with a value range of 0.02-0.04, and λ is the system's own attenuation coefficient, which is obtained by fitting historical data. Step 6.2: Classify the critical warning levels according to the critical confidence level. A confidence level greater than 90% is a Level 1 critical warning, at which point the probability of conflict transformation is extremely high, and control measures need to be taken immediately; 70%-90% is a Level 2 critical warning, at which point conflict transformation is highly likely, and monitoring needs to be strengthened and control measures need to be prepared; 50%-70% is a Level 3 critical warning, at which point conflict transformation is possible, and the trend of conflict evolution needs to be continuously monitored; when the confidence level is below 50%, it is judged as a misjudgment, and data is re-collected for a second judgment. The sampling frequency for the second judgment is increased to 1.5-2 times the original sampling frequency to ensure the reliability of the judgment result. Step 6.3: Generate corresponding state control suggestions based on the critical transformation direction label. These suggestions include the adjustment direction, magnitude, and timing of parameters for both conflicting parties, used to suppress or guide the transformation process. When the transformation direction label is "Party A-led Transformation," the control suggestion is to reduce the strength of Party A's parameters and increase the strength of Party B's parameters. The adjustment magnitude is determined based on the opposition strength value and the correlation weight, and the formula for the adjustment magnitude is: In the formula, Δx is the adjustment range, and k is the adjustment coefficient, with a value range of 0.05-0.

1. When the transformation direction label is "Party B-led transformation", the control suggestion is to reduce the strength of Party B's parameter and increase the strength of Party A's parameter, and the calculation method of the adjustment range is the same as above. When the transformation direction label is "bidirectional mutual transformation", the control suggestion is to adjust the parameters of both parties simultaneously so that the two sides of the contradiction reach a new equilibrium state. The adjustment sequence is determined according to the gradient transfer characteristics, and the parameter of the party with the larger gradient change is adjusted first. After the control suggestion is generated, the predicted result of the contradiction evolution after adjustment is output simultaneously to facilitate the user's evaluation of the control effect.

7. A critical state intelligent determination system based on the transformation of contradictory parties, characterized in that, It includes a multi-source data acquisition module, a data preprocessing module, a contradiction mapping quantification module, a dialectical deduction calculation module, a critical state identification module, a critical output and early warning module, a critical calibration and control module, and a model update module, among which: The multi-source data acquisition module is used to synchronously acquire multi-dimensional physical quantities and state parameters of the target system and construct a time-series data set containing characteristic parameters of both contradictory sides. The data preprocessing module is used to perform normalization and noise reduction preprocessing on the time series data set, remove abnormal disturbance data and retain the true trend information of the evolution of the two contradictory sides; The contradiction mapping quantification module is used to map preprocessed time series data into a quantitative sequence of the opposing forces and a dynamically correlated weight sequence based on preset contradiction opposition mapping rules. The dialectical deduction calculation module is used to perform iterative calculations on the quantified opposition intensity sequence and the dynamic correlation weight sequence using recursive dialectical deduction operators, and to extract the quantitative change accumulation characteristics, gradient transfer characteristics and potential difference decay characteristics of the two sides of the contradiction in the evolution process. The critical state identification module is used to construct a multidimensional critical discrimination feature vector based on the quantitative change accumulation feature, gradient transfer feature and potential difference decay feature, input the multidimensional critical discrimination feature vector into the pre-trained contradiction transformation critical identification model for matching operation, and output the current state interval of the target system. The critical output and early warning module is used to simultaneously output the critical trigger threshold, critical confidence level, critical transformation direction label and graded early warning information when the judgment result is in the critical transformation interval of contradiction; The multi-source data acquisition module specifically includes a distributed heterogeneous sensing unit, a timestamp alignment unit, a data segmentation and slicing unit, a missing value compensation unit, and a data storage unit; wherein: Distributed heterogeneous sensing units consist of various types of sensors; The timestamp alignment unit uses a high-precision time synchronization module to achieve time synchronization of all distributed heterogeneous sensing units with a time synchronization error of no more than 1ms. This unit aligns the data collected by each sensor according to a unified timestamp to form a multi-channel time-series data stream. The data segmentation and slicing unit is used to segment time-series data and set overlapping segments; The missing value compensation unit is used to fill in missing data using a contradictory correlation interpolation algorithm; The data storage unit adopts a distributed storage architecture and consists of multiple storage nodes. The dialectical deduction calculation module includes a recursive operation unit, a historical prior storage unit, a feature extraction unit, an iterative convergence judgment unit, and a deduction log recording unit, wherein: The recursive operation unit is used to perform iterative calculations of the recursive dialectical deduction operator; The historical prior storage unit adopts a high-speed cache design, which can quickly read the inference results of the previous moment as historical prior information; The feature extraction unit is used to extract cumulative quantitative features, gradient transition features, and potential decay features; The iterative convergence judgment unit is used to determine whether the iteration has reached the preset convergence condition and to control the termination of the iteration. The deduction log recording unit is used to record the entire process of iterative calculation, including the time of each iteration, the opposition strength value, the correlation weight value, the deduction output value and the change amount.

8. The system according to claim 7, characterized in that, It also includes a critical calibration and control module, which is used to dynamically calibrate the critical trigger threshold, classify the early warning level according to the critical confidence level, and generate control suggestions for the parameters of the conflicting parties according to the direction of critical transformation. The critical calibration and control module specifically includes a threshold calibration unit, an early warning level classification unit, and a control suggestion generation unit. The threshold calibration unit adopts an adaptive calibration algorithm, which combines the current environmental disturbance intensity and the system's own attenuation coefficient to correct the critical trigger threshold in real time. The early warning level classification unit automatically classifies the critical early warning levels into Level 1, Level 2, and Level 3 based on the level of critical confidence and generates corresponding early warning information. The regulation suggestion generation unit generates targeted parameter adjustment schemes based on the critical transformation direction label and the characteristics of contradiction evolution. The adjustment schemes can be directly output to the regulation execution mechanism of the target system to achieve automatic regulation of contradiction transformation. The system also includes a model update module, which is used to periodically collect new contradiction evolution data to incrementally update the pre-trained model.

9. A computer device, characterized in that, It includes a memory and a processor, wherein the memory is used to store instructions, and the instructions are used to control the processor to perform corresponding operations to execute the intelligent determination method and system for critical states based on the transformation of contradictory sides as described in claim 1.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the intelligent determination method and system for critical states based on the transformation of contradictory sides as described in claim 1.

Images