68 results about "Equivalent transformation" patented technology

The invention provides a low-bit large model quantification method based on sensitivity analysis and abnormal value processing, and the method comprises the steps: dividing a model layer into a high-sensitivity layer, a medium-sensitivity layer and a low-sensitivity layer through employing a hierarchical mixing precision quantification strategy based on sensitivity analysis, and executing different quantification operations; carrying out weight abnormal value suppression and weight layer-by-layer quantization execution by using a quantization method based on two-stage combination; and carrying out activation outlier offline smoothing processing based on equivalent transformation. According to the method, key weights sensitive to quantization errors are identified based on a hierarchical mixed precision quantization strategy of sensitivity analysis, so that a differentiation strategy is adopted in the quantization process, the precision of high-sensitivity weights is reserved, and more aggressive quantization is performed on low-sensitivity weights to minimize model performance loss; the two-stage weight quantization collaboratively solves the problem of abnormal value and error accumulation: the quantization difficulty is migrated from activation to weight offline, the friendliness of activation quantization is improved, outliers do not need to be dynamically detected during operation, and the reasoning speed is prevented from being influenced.

The invention discloses a nonlinear system robust residual error multi-fault detection and reconstruction method based on a sliding-mode observer, and solves the problems that an actuator fault and unknown input disturbance in a servo driving system are closely related and mutually coupled, and a fault detection algorithm is insensitive to identification of a first-degree fault, so that fault report failure or misjudgment is caused. According to the method, a general nonlinear system is transformed into two subsystems with independent fault influence conditions based on global differential homeomorphic transformation for the first time. Equivalently transforming reverse channel disturbance to forward channel disturbance by using a filter, designing a sliding mode state observer according to different actuator fault types, and respectively extracting fault information and unknown input disturbance information of a servo driving system in the two subsystems, thereby realizing decoupling of the fault and the unknown input disturbance. And the sensitivity of fault signal detection and the accuracy and robustness of fault diagnosis are effectively improved.

The invention provides a source database storage process migration method and device, electronic equipment and a storage medium, and the method comprises the steps: carrying out the grammar analysis of an original storage process in a source database, and generating a corresponding abstract syntax tree; based on a preset rewriting rule set, performing semantic equivalent transformation on the flow control syntax structure in the abstract syntax tree to generate a transformed abstract syntax tree; and generating a target script in a target execution environment from the converted abstract syntax tree, executing the target script in a script running engine, dynamically generating a standard SQL statement corresponding to the business logic defined by the original storage process, and submitting the standard SQL statement to one or more target data platforms for execution. And the business logic of the original storage process is reproduced. By introducing the target script as the intermediate execution layer, the flexibility of the migration scheme is greatly improved, so that the migration scheme is suitable for various big data platforms without the native storage process capability.

The invention relates to the technical field of mechanical arm control, in particular to a mechanical arm predictive control method which comprises the following steps: defining the state of a mechanical arm and modeling; collecting data to construct a Koopman linear lifting determinacy prediction model; carrying out probability representation on the error by utilizing Gaussian process regression; constructing a random prediction model by combining a deterministic model and an error model, and deducing a state distribution evolution equation; and converting the joint constraint into probability opportunity constraint, carrying out deterministic equivalent conversion, and solving a rolling optimization problem. According to the method, dependence on a precise dynamic model can be abandoned, precise approximation of nonlinear dynamics is achieved through data driving, high-precision trajectory tracking of the tail end of the mechanical arm is achieved on the premise that it is guaranteed that high probability meets safety constraints, and control conservative property is reduced.

The invention belongs to the technical field of large model quantification, and discloses a large model quantification method based on non-uniform grouping Adamage transform and activation distribution self-adaption, which comprises the following steps: firstly, eliminating abnormal values in an activation matrix through non-uniform grouping Adamage transform, and solving the problem of abnormal distribution caused by uniform grouping; secondly, constructing an equivalent transformation error elimination algorithm, and eliminating a weight quantization error by adopting a GPTQ algorithm; and finally, distributing an optimal quantization data type (Int4 or Nint4) for each activation matrix through an activation distribution adaptive algorithm so as to minimize a quantization error. According to the method, the deployment efficiency of online rotation is improved through non-uniform grouping Adamage transformation, meanwhile, the quantization error is remarkably reduced in combination with self-adaptive data type selection, and the model quantization precision and the hardware deployment efficiency are improved. The method is suitable for compression and optimization of a large language model, and has a wide application prospect.

An electromagnetic transient modeling method and system for a high-efficiency synchronous machine, and, a device. The method comprises: predicting, of a synchronous machine, a first rotor angular velocity, a first rotor angle, and a first current q component of and a first current d component of an armature current, establishing a first Norton circuit for simulating the synchronous machine, and solving at the same time a second Norton circuit after equivalent transformation and a network conductance matrix so as to obtain a three-phase voltage of a synchronous machine port; according to the three-phase voltage, obtaining a second current q component, a second current d component, a second rotor angular velocity and a second rotor angle; and by means of error-controlled iterative solving, determining an electromagnetic transient simulation calculation result of the synchronous machine.

The invention discloses a generation method and device for a mixed precision quantization operator, and relates to the field of deep learning, and the method comprises the steps: converting a quantization program inputted by a user into an initial Qtile calculation graph based on Qtile tensor representation; performing equivalent transformation on the initial Qtile calculation graph to generate a plurality of candidate Qtile calculation graphs which are equivalent to the initial Qtile calculation graph; according to the reuse frequency of operator input data in each candidate Qtile calculation graph, determining an optimal data layout position and an optimal data loading strategy of each candidate Qtile calculation graph; on the basis of the optimal data layout, the optimal data loading strategy and a code generation technology based on a template, codes of all the candidate Qtile computational diagrams are generated, and the codes of the candidate Qtile computational diagrams are achieved through codes of mixed precision quantization operators. According to the invention, an optimized high-performance mixing precision quantization operator can be output.

The invention discloses a model-free adaptive privacy protection control method for a vehicle queue system, and belongs to the technical field of vehicle queue system control and privacy protection. A vehicle virtual output function and a simplified virtual dynamic linear model are established, a homomorphic encryption algorithm based on error learning and a model-free adaptive control method are combined, safe information transmission of a vehicle queue system is ensured, and it is ensured that following vehicles keep a set expected distance and tracking of speeds is consistent; the method only depends on vehicle queue system input and output data, the problem of information loss in a traditional method is effectively avoided, and the model complexity is simplified through equivalent transformation; and meanwhile, a homomorphic encryption algorithm based on error learning is applied to guarantee that the vehicle virtual output information data can realize safe network transmission.

This specification pertains to the field of refining and chemical production processing technology, and particularly relates to a method and apparatus for optimizing refining and chemical production plans for a target secondary processing unit. The method includes: obtaining parameters of the secondary processing unit under a refining and chemical production optimization scenario; constructing a nonlinear constraint equation for product output based on the parameters of the secondary processing unit; performing an equivalent transformation on the nonlinear constraint equation to obtain an equivalent transformation equation; extracting the nonlinear constraint terms and performing a Taylor expansion; transforming the Taylor expansion formula into a linear approximation equation based on the proportion of the feed material to the output material of the blending tank, the feed material quantity to the blending tank, and the feed property values ​​of the blending tank, according to the property balance constraints of the blending tank; and then substituting these into the nonlinear constraint equation for product output, thereby transforming the nonlinear constraint equation for product output into a linear approximation equation containing only material information variables. This method effectively ensures model convergence and shortens the solution time.

The invention introduces a joint optimization technique for task offloading and container caching in containerized edge computing, within the domain of task offloading and container caching. The method involves the steps: constructing a mathematical model based on the containerized edge computing system environment, formulating a joint optimization problem using nonlinear 0-1 programming from the model, and resolving the nonlinear 0-1 programming issue. It establishes a mathematical model for the containerized edge computing system to minimize terminal device task processing time. Additionally, it presents a joint optimization approach for task offloading and container caching. This method transforms the challenging nonlinear 0-1 programming into a solvable linear 0-1 programming problem via an equivalent transformation technique. This addresses the coupling problem between caching and task offloading decisions in edge computing, thereby reducing image file download and container instance startup times in the containerized edge computing system, ultimately shortening terminal device task processing times.

The application provides a water three-dimensional target fusion positioning method using only linear array AOA measurement, only needs to obtain the pitching angle information of the target relative to the array to realize positioning, and greatly reduces the positioning cost in the three-dimensional scene. Compared with the existing estimation method, the SDR estimation avoids the increase of unknown items and constraints caused by regarding the item containing the distance between the target and the array as an unknown quantity in the existing algorithm, and further reduces the calculation amount. Through some mathematical equivalent transformation on the node position error, the influence of the node position error on the positioning performance is reduced as much as possible, and the robust target positioning is realized.

The invention provides a method and a system for analyzing electromagnetic scale dynamic damping characteristics of network construction type equipment, and relates to the technical field of network construction type converters, and the method comprises the steps: carrying out the equivalent transformation of a system energy amplitude dynamic simplified model, and building a system class damping torque model; calculating a system energy amplitude dynamic dominant oscillation mode; based on the system type damping torque model and the dynamic dominant oscillation mode, virtual damping torque contributed by each feedback loop to the system is calculated; according to the virtual damping torque, the relation between the energy amplitude dynamic state and the virtual damping characteristic is established, and therefore the energy amplitude dynamic damping characteristic of the system is judged. According to the method, the oscillation mechanism of the network construction converter under the electromagnetic scale is disclosed, a complete analysis process is provided, and theoretical support is provided for engineering application and control link optimization, so that the oscillation suppression capability and stability of the system are effectively improved.

This invention discloses an ADRC method for attenuating periodic disturbances in time-delay systems, belonging to the field of automatic control technology. The method includes: 1. Obtaining the controlled process model and disturbance frequency, and establishing an extended state model containing the periodic disturbance; 2. Constructing a reduced-order extended state observer and tuning the gain using the bandwidth parameter method; 3. Introducing a pure time-delay module into the observer's output channel to construct an ADRC structure; 4. Revealing the mechanism by which periodic disturbances in time-delay systems cannot be attenuated through two-degree-of-freedom equivalent transformation analysis; 5. Based on the frequency domain characteristic equation, analytically determining the delay of the time-delay module to achieve effective attenuation of periodic disturbances at specific frequencies. This invention solves the problem of attenuating periodic disturbances in time-delay systems without significantly increasing structural complexity. Compared to traditional schemes that require high-gain or complex prediction structures, this invention only requires adjusting the pure time delay of the observer's output channel, and the parameter tuning process is clear and intuitive, significantly lowering the threshold for engineering applications.

The invention belongs to the technical field of chemical engineering, artificial intelligence and model compression, and discloses a uniform quantification method for a large model in the chemical engineering field. On the premise that the expression capability of a model function is not changed, orthogonal equivalent transformation is introduced, combined smoothing processing is carried out on the weight and activation in a linear layer, and activation distribution approximately obeys uniform distribution in a quantization interval by constructing an orthogonal transformation matrix and systematically remodeling the activation distribution, so that the low-bit uniform quantization error is remarkably reduced. And carrying out joint quantization on the smoothed weight and activation by adopting a uniform low-bit uniform quantizer, further introducing a small number of learnable global orthogonal parameters, and carrying out lightweight optimization on module output errors before and after quantization. According to the method, quantitative perception training or large-scale fine tuning is not needed, calculation and storage overhead is low, engineering implementation is simple, inference performance close to full precision can be kept under the 4-bit or 6-bit quantization condition, and video memory occupation and energy consumption in the model deployment process are remarkably reduced.

The application discloses a cross-domain adversarial attack method based on a pre-trained code model, aiming to overcome the shortcomings of existing methods in the challenges of data difference and model feedback loss. The method includes the following steps: obtaining source code data, traversing the source code data to identify its basic constituent unit token, the token at least including an identifier and an operator; generating a natural set of alternative identifiers for the extracted identifiers based on a mask language model; applying token-level equivalent transformation to preliminarily optimize the adversarial examples; applying equivalent structure transformation to further convert the preliminarily optimized adversarial examples to obtain the final adversarial examples. The application enhances the migration ability and attack effect of the adversarial examples through structure and token-level disturbance, combined with the mask language model to generate natural alternative identifiers, and combined with greedy search and gradient optimization.

The present invention relates to the field of communications technology and provides a hybrid precoding method, apparatus, device, medium, and program product. The method comprises: establishing a first problem model of a digital-analog hybrid precoding structure based on an intelligent metasurface (RIS); performing an equivalent transformation on the first problem model to obtain a second problem model; updating matrices and parameters in the second problem model to determine a digital-analog hybrid precoding matrix; and performing hybrid precoding on a transmitted signal based on the digital-analog hybrid precoding matrix. The present invention replaces traditional antennas and phase shifters with low-power, low-complexity RIS to achieve low-complexity, energy-efficient coding, thereby reducing power consumption and hardware costs.

The invention provides a smooth regularization method and device for power descent guidance with path constraint, and the method comprises the steps: building an optimal control model of lander fuel optimal power descent guidance under a three-dimensional rectangular coordinate system with a planetary landing point as an original point, carrying out the equivalent transformation of the optimal control model through unified smooth regularization, and carrying out the smooth regularization of the optimal power descent guidance. And introducing a Hamiltonian function based on the Pontryagin minimum principle to determine an optimality condition of unified smooth regularization, and finally performing iterative solution on the optimal control model after equivalent transformation based on the Hamiltonian function and the optimality condition to obtain the optimal control quantity of the lander in the power descent stage. By the adoption of the method, the problems that bang-bang control behaviors exist in an existing lander power descent guidance technology, and inequality path constraints are difficult to process effectively can be solved.

The invention discloses a downlink communication rate optimization method and system based on a clip antenna array, and a storage medium, a base station provides services for a single antenna user randomly distributed in a square with a side length D through a plurality of clip antennas, and the method comprises the steps: building a rate model based on a finite code length theory, constructing a clip antenna position optimization problem, an objective function with the maximum user rate as the optimization objective is obtained, the optimization variable is the abscissa of the position of the clip antenna, and the optimization constraints are the minimum antenna spacing constraint and the QoS constraint; equivalently converting the QoS constraint into a signal-to-noise ratio constraint, and carrying out equivalent transformation on the target function; deriving a closed-form solution of the antenna position; and adjusting the closed-form solution of the antenna position according to the phase constraint to obtain the optimal solution of the antenna position optimization problem. Complex radio frequency link configuration in a traditional array is avoided, and the reachable rate under the limited code length is effectively improved under the condition that the service quality and the antenna spacing constraint are guaranteed.

The invention relates to the technical field of computers, and provides a floating point multiplication and addition instruction fusion method and device, equipment and a storage medium. According to the method, a fusion scaling factor and a fusion offset of two continuous floating point multiplication and addition operations are calculated, the two continuous floating point multiplication and addition operations are merged based on the fusion scaling factor and the fusion offset, and a fusion instruction is output. And two FMA operations are combined into a single instruction by utilizing mathematical equivalent transformation, so that the problems of relatively large computing load, relatively high computing delay and low hardware utilization rate of a GPU (Graphic Processing Unit) caused by relatively large calling of floating point multiply-add instructions in exponential approximate calculation in existing softmax are solved.

The present invention provides a multi-agent MEMG scheduling method based on distributed robust optimization, the method comprising: constructing a multi-agent MEMG-prosumer operation model with MEMG as the upper layer and prosumers as the lower layer; constructing an objective function minCoM for a single-agent MEMGi based on minimizing its own operation cost; MGi ; The minCoM MGi represents the minimum value of the total operating cost of MEMGi; optimizes the multiple uncertain subjects in the multi-agent MEMG-prosumer operation model based on the Wasserstein sphere; constructs the objective function of the multi-agent MEMG microgrid alliance based on the Nash negotiation theory; constructs the objective function PR for the prosumer J within MEMGi based on its own minimization operation cost model i,J , the PR i,J represents the total cost of the Jth prosumer within the MEMGi; the MEMG-prosumer model is equivalently converted and solved. This invention improves the overall operational economics of multi-agent MEMGs through cooperative game-playing.

The application provides a heat dissipation structure design method of an intelligent processor, comprising the following steps: a model reconstruction step, combining a design process of a structure design software and a thermal simulation software, reconstructing models according to modeling results of the two design processes, and obtaining temperature results of the two design processes; an error calculation step, the thermal simulation software calculating error parameters of the two design processes according to the temperature results of the two design processes; and an equivalent transformation step, the structure design software performing equivalent transformation design according to the error parameters. The application also provides a heat dissipation structure design device of an intelligent processor, a storage medium and an electronic device. Therefore, the application fully utilizes the advantages of the two heat dissipation structure design methods, simplifies the process, reduces the design cost, and improves the design efficiency.

The application discloses a ship resilience reinforcement course keeping control method under network attack, comprising the following steps: establishing a responsive nonlinear Nomoto model as a ship motion mathematical model to obtain a simplified transfer function; constructing a ship course robust controller according to the simplified transfer function based on a second-order closed-loop gain shaping algorithm; obtaining a ship course error under consideration of ship network attack to perform nonlinear modification on the control input of the ship course robust controller; obtaining the control input of the ship course robust controller after nonlinear modification, and performing equivalent transformation model of input signal feedback; optimizing the control output of the ship course robust controller according to the equivalent transformation model by using a zero-order holder to obtain optimized rudder angle control output; and obtaining a final ship navigation course according to the optimized rudder angle control output in consideration of marine environmental interference. The technical problem that the traditional ship control strategy cannot resist network attack to guarantee the ship course and collision avoidance control functions is solved.

The application provides a method for data collection and unloading of distributed nodes in an unmanned aerial vehicle (UAV) assisted Internet of Things (IoT), and belongs to the technical field of UAV assisted mobile communication networks.The method is used for solving the joint data collection and computing unloading task of large-scale IoT nodes and realizing energy efficiency maximization in an UAV assisted mobile edge computing (MEC) system.The application is attributed to a nested mixed integer optimization problem, wherein an external problem is to select a specific number of key INs from potential INs for data transmission, and an internal problem is to maximize the energy efficiency of the UAV by optimizing the scheduling, three-dimensional flight trajectory and computing resource allocation of the UAV so as to perform data collection and computing unloading from the selected INs.The external problem is solved by using a method based on Gale circle alignment to convert the selection of INs into a matrix eigenvalue optimization problem, the external problem is divided into three sub-problems, and the model is convexized by using equivalent transformation, convex difference and continuous convex approximation to solve the problem.

The invention discloses a push-pull class-E inverter with a low-harmonic constant-current output characteristic and a design method, provides a novel class-E push-pull circuit topological structure, and redesigns the position and element of an output slot based on a traditional series stacked push-pull structure; a corresponding inverter design method is used for carrying out equivalent transformation on the circuit to obtain an equivalent circuit of the two E-type resonance units, and a circuit model is simplified; establishing a mathematical model of a single E-type resonance unit according to the equivalent circuit model, and solving a circuit through a load-independent constant current condition and a soft switching condition to obtain a mathematical solution meeting the load-independent condition; the working mechanism of the push-pull structure formed by the two E-type resonance units is analyzed, the low harmonic characteristic of the output current is disclosed, the inverter parameter design process and method are determined, and finally an inverter circuit physical model is established through parameter design. According to the invention, a good constant current output characteristic can be maintained without closed-loop control, and a low harmonic output characteristic can still be maintained without an output filter.

The present invention discloses a method and system for detecting and optimizing the oscillation of strongly connected components in a combinational logic circuit. The method constructs an oscillation detection module based on the external input signals and their signal values in the strongly connected components, and then constructs an equivalent transformation module in combination with a multiplexer. The equivalent transformation module is inserted into the output end of the common logic gate of the nested logic loop containing a negative feedback loop and the modified netlist is output. The present invention can successfully detect and interrupt the oscillation when the circuit oscillates, and does not affect the original circuit logic function when the circuit does not oscillate, improving the simulation efficiency and stability of the combinational circuit.

This invention belongs to the field of wireless communication, specifically relating to a STAR-RIS-assisted 6G integrated sensing and communication beamforming design method. The method includes constructing a STAR-RIS-assisted integrated sensing and communication system model; under constraints of transmit power, sensing performance, and multi-user service quality, establishing an optimization problem aimed at maximizing multi-user communication and speed by jointly designing the base station transmit beamforming and the transmission-reflection coefficients of STAR-RIS; transforming the optimization problem into an easily tractable equivalent form based on a weighted minimum mean square error equivalent transformation; decoupling the optimization problem based on the equivalent form into two independent sub-problems using an alternating optimization method; solving the sub-problems using a successive convex approximation method and a semi-definite relaxation method, and iteratively updating the algorithm until convergence; this invention effectively improves system communication performance while ensuring sensing performance and multi-user communication quality.

The invention relates to the technical field of computers, and discloses an intelligent contract compiler defect detection method and a medium, and the method comprises the steps: obtaining an intelligent contract source code; performing reverse optimization equivalent transformation on the source code to generate a variant contract; inputting the variant contract into the same compiler, and compiling in a non-optimization mode and an optimization mode respectively to obtain a first byte code and a second byte code; comparing static characteristics and runtime behaviors of the first byte code and the second byte code; and when inconsistency exists, outputting a compiler defect report. According to the method, three core technical principles of reverse optimization transformation, single compiler multi-level difference and static and dynamic two-dimensional verification are adopted. The hidden defect of how to effectively detect the optimization stage of an intelligent contract compiler under the scene that the intelligent contract compiler only has a single mainstream implementation is solved, and meanwhile, the limitation that a traditional random difference test depends on multiple compilers is broken through.

The embodiment of the invention discloses a method for determining the normal section bending moment of an adapter ring of a steel-concrete hybrid wind power tower drum under an axial pressure working condition. The method comprises the following steps: acquiring a preset axial pressure on the adapter ring, a preset prestress generated by a prestressed tendon and a geometric dimension of the adapter ring; according to a preset stress equivalent conversion scheme, the adapter ring of any arc angle in the adapter rings is taken as an isolator for stress analysis, and target parameters are obtained through simplified calculation based on the axial pressure, the prestress and the geometric dimension; and obtaining the normal section bending moment of the adapter ring through a pre-established normal section bending moment formula according to the target parameter. According to the invention, the blank of the existing design specification is filled; through theoretical derivation and finite element verification, the reliability of a calculation result is ensured; and a scientific basis is provided for optimization design and safety evaluation of the adapter ring, and the engineering risk is reduced.

The invention discloses a tracked robot trajectory tracking control method based on a dynamic virtual steering coefficient, which relates to the technical field of tracked robots, and comprises the following steps: randomly selecting an expansion point on a reference trajectory to carry out Taylor expansion on a kinematics equation of a tracked robot to obtain a linear approximation equation; wherein a dynamic virtual steering coefficient is added to the kinematics equation; sequentially obtaining a linear error equation, a discretized kinematics model, a state recursion formula and a relational expression of a state quantity and a control quantity in a prediction time domain based on the linear approximation equation; performing equivalent transformation on the cost function by using the relational expression of the state quantity and the control quantity in the prediction time domain to obtain an equivalent cost function; the equivalent cost function is converted into a standard quadratic programming form and then is solved, and an optimal control quantity matrix in a prediction time domain is obtained; updating the dynamic virtual steering coefficient based on the position error and the average speed in the prediction time domain; according to the invention, the trajectory tracking precision and the adaptive capability can be improved.

The application provides a mechanical system analysis and experiment method based on circuit analogy, which provides two kinds of circuit analogy methods for two kinds of mechanical systems of rotation and translation, can satisfy simulation of mechanical systems with different complexity, specifically, variables in the circuit are analogous to variables in the mechanical system, the mechanical system is converted into an equivalent circuit, the equivalent circuit is analyzed by using a circuit analysis method, and an analysis result of the mechanical system can be directly mapped to obtain an analysis result. In terms of analysis and calculation, the circuit calculation method is mature, loop current method and node voltage method can be used to solve the circuit steady state; the circuit theorem is complete, the equivalent transformation principle is clear, the circuit can be simplified; the circuit time-frequency domain analysis method is mature, the network function method can be used to analyze the small signal dynamic characteristics; the circuit element has no rotating part, the variable direction is easy to understand and represented in a plane. In terms of experiment, the circuit element is rich in types and easy to obtain; the circuit element parameter is easy to measure, and the element with the required parameter can be obtained through series and parallel connection.