38 results about "Automatic differentiation" patented technology

The disclosed system provides a transformation-based implementation of forward-mode and reverse-mode automatic differentiation as a built-in, first-class function in a functional programming language. Each of these constructs imposes only a small constant factor of the computational burden (time) of the function itself, and the forward construct has the same properties in terms of space. The functions can be applied to any function, including those involving derivatives and nested closures.

The invention discloses a distributed power system state estimation method. A distributed power system state estimation model is established in consideration of distributed generation (DG) characteristics so that the real-time accurate operational state of each node in the power grid can be obtained, and the real-time accurate operational state of the DG can also be obtained; secondly, a preserving non-linear method is introduced to the state estimation, and the truncation error is avoided so that the state estimation is provided with a higher computational accuracy and the iteration efficiency of an algorithm is improved; and finally, the Jacobian matrix and the Hessian matrix of a function is calculated and measured in the preserving non-linear method of the state estimation, the work load of manually calculating a large quantity of differential functions and compiling differential codes is large, and the work is too cumbersome and easy to produce errors so that the distributed power system state estimation method utilizes the automatic differentiation technique to replace the traditional manual compiling differential codes to calculate the Jacobian matrix and the Hessian matrix, the workload of the manual compiling codes is reduced, and the development efficiency of procedures is improved.

A method and associated algorithms for identifying and distinguishing geometric feature signals from defect signals in the NDE of longitudinal structures. The method includes the steps of collecting an interrogation signal (including reflected components) from a longitudinal structure under evaluation and comparing it with a selected reference signal from a known geometric feature maintained in a database. The comparison involves a determination of the signals phase. Same phase signals identify the source as a geometric feature, while opposite phase signals identify the source as a defect. The comparison involves the steps of gating each of the signals and creating an array of correlation values between points on each. The correlation values are analyzed and a determination (based on comparing maximum and minimum correlation values) is made of the signal phases. A reliability factor may be determined by comparison of the correlation values and the maximum and minimum thereof.

The invention discloses a power system flow optimization method for effectively processing a complex electric power control device. The method adopts a node power injection method for modeling for the electric power control device, performs processing to the device and enables the device to be integrated to the traditional flow optimization by using a superposition principle, and effectively calculates the corrected matrixes of the Jacobian matrix and the Hessenberg matrix of the electric power control device to the objective function and the constraint condition by using an automatic differentiation technology. The method is efficient, flexible and feasible, keeps the high efficiency of the traditional flow optimization procedure, protects the existing code investment, can extend to other user-defined models and meets the requirements of analysis, running and planning of the modern power system.

The invention discloses a voltage source commutation-high voltage direct current (VSC-HVDC) power flow computing method based on automatic differential (AD) and a reserving non-linear method, which applies the AD technology to solution of the problem of alternating-direct current power flow containing VSC-HVDC through the reserving non-linear method. The AD is utilized to automatically generate Jacobian matrix of a power flow equation. Compared with manual programming, the method reduces handwriting code quantity, improves developing efficiency of programs. Further, the Jacobian matrix of the method is more constant, so that iteration times are increased compared with the traditional unifying iteration method. The AD only needs to be called once to form the Jacobian matrix, and program operation time is greatly reduced. A plurality of computing examples verify effectiveness of the method in solving alternating-direct current power flow problems. The results of the computing example prove that the reserving non-linear method is higher in speed than the traditional unifying iteration method under the same conditions.

The invention discloses a power system harmonic state estimation method based on an automatic differentiation. The automatic differentiation (AD) technology is applied to the power system harmonic state estimation. The method includes that firstly storage allocation is performed for the AD, live variables are described, a network parameter is obtained, a harmonic number is set, and a node admittance matrix of a harmonic is formed; then before iteration, positions and values of invariant elements in a jacobian matrix are stored in a linked list; and the iteration starts, the AD tool is used to calculate variable elements in the jacobian matrix, and simultaneously the invariant elements of the jacobian matrix in the linked list are read to obtain the jacobian matrix required by the harmonic state estimation. According to the power system harmonic state estimation method based on the automatic differentiation, traditional manual differential code writing is replaced by the AD to calculate the jacobian matrix, thereby handwritten code quantity is reduced, development efficiency of programs is increased, truncation error is effectively avoided, computational efficiency of algorithm is improved, and the power system harmonic state estimation method based on the automatic differentiation is convenient to implement on an existing state estimation software.

The invention relates to the field of natural language processing, and in particular relates to a construction method of a domain dictionary. The method comprises the following steps: on the basis of automatic acquisition of a text keyword, clustering to-be-processed texts to form different topic text sets; selecting a part of seed words in a domain text set of a to-be-constructed dictionary through manual examination; on the basis, analyzing the distance of the clustered topic text sets and the selected domain seed words in relationship, and only retaining the top text sets which are relatively close in relationship for expanding the domain dictionary; and in a related domain, performing automatic expansion of the domain dictionary by combining an algorithm in the related domain to obtain a corresponding dictionary. According to the method provided by the invention, the to-be-constructed domain dictionary can be automatically expanded through a few of part of seed words on the basis of automatic differentiation of the domains of the text topics; the construction efficiency of the dictionary is relatively high, the accuracy is high, and the pertinence of the domain is strong; the method has wide application prospect in text analysis and natural language processing field.

The invention provides a design method of a position and attitude observer for an airborne parallel platform. According to the design method of the position and attitude observer for the airborne parallel platform, a six-degree-of-freedom parallel machine platform serves as the airborne platform, the position and attitude of a fixed end of the platform dynamically changes along with a body, and the position and attitude of a workbench which is controlled by six parallel extensible rods and universal hinges is always located in the horizontal state; the actual length of the extensible rods is measured through a sensor and the position and attitude of the platform is estimated through a nonlinear observer; a measurement error of the length of the extensible rods serves as an input signal of position and attitude estimation through a state observer; the nonlinear observer is transformed into a linear observer with a nonlinear perturbation through differential homeomorphism for calculation; meanwhile the estimation of a gain matrix is achieved by an automatic differential technology and accordingly the solving speed of the gain matrix is greatly improved.

The invention discloses an ICT turntable test machine. The test machine comprises an equipment frame, a workbench component, a CCD (Charge Coupled Device) component, a test moving component, and X-axis material receiving component, a turning component and a swing plate unloading component, wherein the workbench component is rotatably arranged on the table top of the equipment frame; the test moving component, the X-axis material receiving component, the turning component and the swing plate unloading component are arranged on the table top of the equipment frame; and the CCD component is arranged on the test moving component. Through the cooperation of the test moving component, the X-axis material receiving component, the turning component, the swing plate unloading component and the CCD component, automatic position alignment, automatic test, automatic differentiation of qualified products and unqualified products, and automatic swing plate unloading can be realized, the equipment is simple in operation, the labor intensity of workers is reduced by replacing manual operation by a robot, the production efficiency of ICT test is improved, and the risk of test leakage or material mixing of an ICT product is reduced.

A method and an apparatus allow learning a program that is characterized by a set of parameters. In addition to carrying out operations of the program based on an input vector and the values of the parameters, the method also carries out automatic differentiation steps over the operations of the program to compute derivatives of the output vector with respect to the parameters to any desired order. Based on the computed derivatives, the values of the parameters of the program are updated.

The invention discloses equipment for integrating full-automatic electrical testing, feeding, cutting and marking of an FPC bare board. The equipment includes a sheet metal equipment cabinet and an internal functional module arranged in the sheet metal equipment cabinet. The internal functional module includes a first installation plate, a second installation plate and a third installation plate.A feeding mechanism, an XY stepping mechanism, a carrier plate rotating mechanism and a press mechanism are installed on the first installation plate, an OK/NG/marking plate moving mechanism is installed on the second installation plate, a paper suction Z-axis mechanism and a marking mechanism are installed on the third installation plate, and the internal functional module further includes an X-axis material handling mechanism. The equipment can effectively achieve FPC bare board automatic testing, automatic feeding and cutting, automatic paper placement and automatic differentiation of OK/NGproducts, and integrates an automatic marking function and an automatic marking identification function, the occupied space of the equipment is reduced, labor in FPC bare board automatic testing is saved, the yield of defective products caused by manual identification is reduced, the test efficiency is improved, and the stability of the equipment is better.

The invention discloses a beam adjustment method hardware accelerator based on a Zynq FPGA. Shuer elimination, jacobian matrix updating and cost function calculation parts in a beam adjustment methodare realized at a PL end of Zynq through circuit design, and other parts are realized at a PS end of Zynq through software programming; an accelerator PL end splits two hardware circuits according toan algorithm, independent hardware is used for Shuer elimination, and Jacobian matrix updating and cost function calculation share the same hardware; wherein different computing units are formed in aninternal circuit of the accelerator, data transmission is carried out between the computing units by adopting a dual-port RAM (Random Access Memory) on an FPGA (Field Programmable Gate Array) chip, afinite-state machine is adopted in each computing unit, and read-in, read-out and storage of each step of data are controlled through state conversion; the Jacobian matrix updating and cost functioncalculating module calculates a Jacobian matrix by adopting a mode of combining analytic differential and automatic differential and calculates a cost function at the same time. BA calculation energyconsumption can be greatly reduced.

There is provided a method and corresponding systems and computer-programs for evaluating and/or adapting one or more technical models related to an industrial and/or technical process. The method comprises obtaining (S1) a fully or partially acausal modular parameterized model of an industrial and/or technical process comprising at least one physical sub-model and at least one neural network sub-model, including one or more parameters of the parameterized model. The method further comprises generating (S2) a system of differential equations based on the parameterized model, and simulating (S3) the dynamics of one or more states of the industrial and/or technical process over time based on the system of differential equations. The method also comprises applying (S4) reverse-mode automatic differentiation with respect to the system of differential equations when simulating the industrial and/or technical process in order to generate an estimate representing an evaluation of the model of the industrial and/or technical process.

The invention provides a temperature state estimation method based on automatic differentiation (AD). In a traditional state estimation calculation process, the line resistance is always invariant. The transmission line resistance changes with the change of external temperature. According to the invention, temperature used as a new state quantity is introduced into the state estimation process to establish a state estimation model considering temperature changes; in addition, a Jacobian matrix for calculating a measurement function is required in the state estimation; since a branch temperature variable is taken into account, massive differential functions are manually calculated and a differential code is written, which is too cumbersome and error-prone; AD is used to replace the traditional manually-calculated Jacobian matrix, which reduces cumbersome manual coding and error probability and effectively avoids a truncation error; the efficiency of program development is improved; and the method is easily combined with the existing state estimation software.

The embodiment of the invention discloses a quantum control waveform optimization method and device, computer equipment and a storage medium. The method comprises the steps of performing Fourier transform on a preset initial waveform of a quantum logic gate to obtain frequency domain data of the preset initial waveform; performing bandwidth limitation on the frequency domain data according to a preset bandwidth limitation condition to obtain target frequency domain data; performing inverse Fourier transform on the target frequency domain data to obtain a target waveform; performing quantum dynamics evolution on the target waveform, and calculating a loss function of a quantum logic gate; performing automatic differentiation on the target waveform by using the loss function through a neural network to obtain a gradient of the loss function; judging whether an end condition is met or not according to the loss function and the gradient; if not, updating the preset initial waveform by using a gradient descent method based on the target waveform, and repeating the method; and if so, determining the target waveform as an optimized waveform and outputting the optimized waveform. Therefore, the quantum control waveform which is controllable in frequency domain, smooth and analyzable is obtained.

The invention relates to the field of natural language processing, in particular to a domain dictionary generation system for natural language processing. On the basis of automatic differentiation of text topic domains, the system conducts the automatic construction of dictionaries of corresponding domains based on seed words. A user inputs the to-be-processed texts and domain seed words into the system. On the basis of the automatic acquisition of the test key words, the to-be-processed texts are clustered, and relationship level between the clustered thematic text set and the selected domain seed words is analyzed and decided. The automatic expansion of the dictionaries is conducted with coordination of algorithms in the thematic text set with close relationship. According to the domain dictionary generation system for natural language processing, on the basis of the automatic differentiation of text topic domains, the automatic expansion of domain dictionaries is achieved through small quantity of seed words. The dictionary construction efficiency is high, the accuracy is high, and the domain pertinence is strong. The domain dictionary generation system for natural language processing provides a powerful tool for the text analysis and natural language processing.

The invention discloses a method for calculating porous medium oil-water two-phase flow based on windward GFDM. The method comprises the following steps: S1, establishing a porous medium oil-water two-phase flow model; S2, discretizing the porous medium oil-water two-phase flow model based on windward GFDM (Generalized Frequency Division Multiplexing) to obtain a fully implicit discretization format of the porous medium oil-water two-phase flow model; and S3, solving the fully implicit discrete format of the porous medium oil-water two-phase flow model by using a nonlinear solver based on Newton iteration and automatic differentiation to realize high-precision meshless calculation of pressure and water saturation. The invention provides a method for calculating porous medium oil-water two-phase flow based on windward GFDM (Generalized Fused Deposition Modeling), and provides a reference basis for forming a porous medium flow universal meshless numerical simulator independent of a grid.