CN111310113B - Counter example generation method and device of expert rule system based on time sequence data - Google Patents
Counter example generation method and device of expert rule system based on time sequence data Download PDFInfo
- Publication number
- CN111310113B CN111310113B CN202010090753.1A CN202010090753A CN111310113B CN 111310113 B CN111310113 B CN 111310113B CN 202010090753 A CN202010090753 A CN 202010090753A CN 111310113 B CN111310113 B CN 111310113B
- Authority
- CN
- China
- Prior art keywords
- sequence
- generating
- expert
- inverse function
- control parameter
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/14—Fourier, Walsh or analogous domain transformations, e.g. Laplace, Hilbert, Karhunen-Loeve, transforms
- G06F17/141—Discrete Fourier transforms
- G06F17/142—Fast Fourier transforms, e.g. using a Cooley-Tukey type algorithm
Abstract
The invention provides a counter example generation method and a counter example generation device of an expert rule system based on time sequence data, wherein the method comprises the following steps: acquiring an expert rule of an expert rule system based on time sequence data of industrial equipment; and outputting the original sequence of the time sequence data by adopting a counter example generation module according to the expert rule. The scheme of the invention can realize the accurate early warning of the common time sequence data in the industry.
Description
Technical Field
The invention relates to the technical field of information processing of industrial equipment, in particular to a method and a device for generating counter examples of an expert rule system based on time sequence data.
Background
In the data processing of an industrial plant, expert rules are exemplified by the following table:
in the above table, for example, the content of the column corresponding to the thermal bending of the rotor of the industrial equipment is a sign of the rotor; the row corresponding to the mechanical or electrical deviation of the journal is a sign of the journal; the column corresponding to the poor dynamic balance is the sign corresponding to the dynamic balance.
In the prior art, many expert rules are not precise, for example, there is a main energy of 2Hz in the rotation speed of the fan, and what frequency domain is "2 Hz"? To what extent the 2Hz energy is large can be referred to as "main energy"?
In the prior art, no ready "label" data exists to check or train expert rules; a simple idea: by automatically generating a certain number of counter cases (aiming at the current threshold value setting), the expert is enabled to mark the label in a targeted manner, and the expert rules are guided to be refined. But randomly generated cases, either with too low a negative proportion or too far from the positive, do not help to refine the threshold.
Disclosure of Invention
The invention provides a counter example generation method and device of an expert rule system based on time sequence data, which can realize accurate early warning of common time sequence data (such as sensor data) in the industry.
In order to solve the above technical problem, an embodiment of the present invention provides a method for generating a counter example of an expert rule system based on time series data, including:
acquiring an expert rule of an expert rule system based on time sequence data of industrial equipment;
and outputting the original sequence of the time sequence data by adopting a counter example generation module according to the expert rule.
Optionally, according to the expert rule, outputting the original sequence of the time sequence data by using a counter-example generation module, includes:
inputting the expert rules into an oriented acyclic graph parser of the expert rules for parsing to obtain an oriented acyclic graph of the expert rules;
finding out all path sequences from a first-stage node to a last-stage node in the directed acyclic graph;
on the last-stage node of one of the channel sequences, inputting the transformed sequence output by the first time sequence characteristic function into a first characteristic inverse function corresponding to the node to obtain a first sequence; the first characteristic inverse function is an inverse function of the first timing characteristic function;
inputting the first sequence into a second characteristic inverse function corresponding to a previous node of a last node of the directed acyclic graph to obtain a second sequence until the directed acyclic graph outputs an original sequence of the time sequence data; the second characteristic inverse function is an inverse function of the second characteristic function corresponding to the previous-stage node.
Optionally, outputting the original sequence of the time series data by the directed acyclic graph includes:
randomly generating a plurality of groups of control parameter combinations according to a key control parameter list of a given judging rule; the control parameter list is the control parameter of the inverse function of the feature;
and selecting one group of control parameter combinations from the plurality of groups of control parameter combinations, and generating corresponding original sequences by utilizing the inverse function of the characteristics.
Optionally, selecting a group of control parameter combinations from the plurality of groups of control parameter combinations, and generating a corresponding original sequence by using a characteristic inverse function, including:
generating a plurality of groups of expected index combination sets at the current numerical value according to the size of the sliding window, the slope threshold and the proportion threshold;
randomly generating a curve slope data set of a slope threshold value and a ratio threshold value combination value which accord with a judging rule according to the size m of a specified input array;
and generating a corresponding original sequence according to the curve slope data set.
Optionally, generating a corresponding original time sequence according to the slope array with the length of m includes:
and in each time window, generating a standard straight line according to the slope in the slope array with the length of m, and then generating a corresponding original sequence after scrambling.
Optionally, selecting a group of control parameter combinations from the plurality of groups of control parameter combinations, and generating a corresponding original sequence by using a characteristic inverse function, including:
determining the position of the frequency in the Fast Fourier Transform (FFT) array according to the size of the sliding window and the rotating speed;
randomly generating a phase according to the amplitude of the frequency, and determining the numerical value of the phase in the FFT array;
for the rest positions, randomly generating complex numbers;
and performing inverse FFT on the FFT array to obtain a corresponding original sequence.
Optionally, selecting a group of control parameter combinations from the plurality of groups of control parameter combinations, and generating a corresponding original sequence by using a characteristic inverse function, including:
generating a standard rolling curve from a plurality of historical rolling processes;
randomly selecting a plurality of wave crest areas according to the times of wave lack heads;
generating a plurality of flat wave head fragments with different lengths near the wave crest according to the minimum length and the maximum change rate;
and replacing the original wave head with the generated flat wave head to obtain a corresponding original sequence.
Optionally, the method for generating a counter example of the expert rule system based on the time series data further includes:
and performing cross processing among cases and/or variation processing of the cases on the generated original sequence, and outputting.
Optionally, the method for generating a counter example of the expert rule system based on the time series data further includes:
and performing at least one of marginal degree evaluation, disturbance and case screening on the output original sequence to obtain a final output sequence.
The embodiment of the present invention further provides a counter example generating device of an expert rule system based on time series data, including:
the acquisition module is used for acquiring expert rules of an expert rule system based on the time sequence data of the industrial equipment;
and the processing module is used for adopting a counter example generation module according to the expert rule and outputting the original sequence of the time sequence data.
The technical scheme of the invention has the beneficial effects that: according to the embodiment of the invention, the original sequence of the time sequence data is output by adopting a counter-example generation module according to the expert rule; greatly accelerates the precipitation process of expert experience and knowledge. Machine learning, artificial intelligence, etc. techniques rely on a large number of data samples, while in industrial or commercial operations, anomalous samples are very rare (e.g., blade fractures are rarely actually occurring). The expert experience often lacks quantification, no method is available for automatic application, and the embodiment of the invention generates samples with special pertinence (rather than randomly), and the generated cases can continuously approach the limit of the expert experience, thereby accelerating the quantification and the precision of the expert knowledge.
Drawings
FIG. 1 is a flow chart of a counter example generation method of an expert rule system based on time series data according to an embodiment of the present invention;
FIG. 2 shows a schematic diagram of an industry expert rule in a rules engine towards an acyclic graph DAG;
FIG. 3 is a schematic diagram illustrating an embodiment of a counter-example generation system of an expert rule system based on time series data according to the present invention;
FIG. 4 is a process flow diagram of a rule DAG parser in an embodiment of the invention;
FIG. 5 is a schematic diagram illustrating the processing flow of an inverse function execution engine in an embodiment of the present invention;
FIGS. 6, 7 and 8 are diagrams illustrating an exemplary process of generating an original sequence and a correlation sequence in the case of expert rules with ascending trends in variables in embodiments of the present invention;
FIGS. 9 and 10 are schematic diagrams of a variation timing and correlation sequence for generating 1X amplitude in an embodiment of the present invention;
FIGS. 11 and 12 are schematic diagrams showing the timing of the change in the number of missing headers and the correlation sequence;
FIG. 13 is a schematic diagram of case variations in an embodiment of the present invention;
FIG. 14 shows a schematic diagram of a security cross in an embodiment of the invention;
fig. 15 shows a schematic diagram of boundary case screening in an embodiment of the invention.
Detailed Description
In order to make the technical problems, technical solutions and advantages of the present invention more apparent, the following detailed description is given with reference to the accompanying drawings and specific embodiments.
As shown in fig. 1, an embodiment of the present invention provides a method for generating a counter example of an expert rule system based on time series data, including:
and step 12, adopting a counter example generation module according to the expert rule, and outputting the original sequence of the time sequence data.
In the embodiment of the invention, the original sequence of the time sequence data is output by adopting the counter example generation module according to the expert rules aiming at the time sequence data (such as sensor data common in the industry) of the industrial equipment instead of general data (such as text data and multimedia data), so that the counter examples are accurately generated by using the characteristic 'boundary' characteristic in the rule condition instead of randomly generating the counter examples.
In an alternative embodiment of the present invention, step 12 may include:
step 121, inputting the expert rules into an expert rule directed acyclic graph analyzer for analysis to obtain an expert rule directed acyclic graph;
step 122, finding out all path sequences from the first-stage node to the last-stage node in the directed acyclic graph;
step 123, inputting the transformed sequence output by the first time sequence feature function into a first feature inverse function corresponding to the node on the last-stage node of one of the path sequences to obtain a first sequence; the first characteristic inverse function is an inverse function of the first timing characteristic function;
step 124, inputting the first sequence into a second inverse characteristic function corresponding to a previous node of a last node of the directed acyclic graph to obtain a second sequence until the directed acyclic graph outputs the original sequence of the time sequence data; the second characteristic inverse function is an inverse function of the second characteristic function corresponding to the previous-stage node.
Specifically, as shown in fig. 2, an example of how the industrial expert rules are expressed in the rule engine to the acyclic graph DAG is shown, in which M represents a raw Measurement (Measurement), D represents a feature variable (Descriptor), S represents a Symptom variable (Symptom), and F represents a judging logic expression. OP _ D represents a characteristic (Descriptor) Operator (Operator), OP-S represents a Symptom (Symptom) Operator, and Par represents a control parameter corresponding to the Operator. In the directed acyclic graph, in a normal information processing procedure, M is input to OP _ D (a feature operator, such as the above-mentioned feature function), a feature variable D is output after processing, the output feature variable D is input to OP-S (a symptom operator, such as a symptom function), a symptom variable is output after processing, these symptom variables are input to F (a study and judgment logical expression), and an expert rule is finally output.
For example, M1 to M6 indicate state monitoring amounts at different points of the rotary machine, for example, M1 indicates an X-direction vibration amount of an air inlet end, M2 indicates a Y-direction vibration amount of the air inlet end, M3 indicates a rotation speed, M4 and M5 indicate vibration amounts of an air outlet end X, Y, and M6 indicates a machine casing vibration amount. OP _ D1 represents the 1X multiplier magnitude calculation operator, D1_1 is the 1X multiplier magnitude corresponding to M1, and the sliding window size of FFT calculation represented by Par 1. OP _ D5 represents the constituent phase map for the X, Y direction; OP-S1 represents the "1X multiplier amplitude rapidly getting larger" feature operator, and par7 is the slope' S decision threshold parameter.
As shown in fig. 3, in an embodiment of the present invention, a specific architecture diagram of a counter-example generating module is shown, and the counter-example generating module includes: the system comprises a rule DAG resolver, an inverse function execution engine, an inversion task engine and a control parameter combination generation engine;
as shown in fig. 4, the rule DAG parser is configured to parse the input expert rule to obtain a directed acyclic graph of the expert rule; according to DAG (directed acyclic graph) generated by a rule, adopting a reverse propagation mode to gradually generate an original variable, for example, according to M output by a judging logic expression F, inputting the M into an inverse function of the characteristic function, and finally outputting a sequence V; it should be noted that it is not necessary to implement the "end-to-end" full-range inversion, but some stages of inverse functions may not be implemented or may be implemented with great difficulty.
As shown in fig. 5, the inverse function execution engine is configured to input, on a node at the last stage of one of the path sequences, a transformed sequence output by the first timing feature function into the first feature inverse function corresponding to the node, so as to obtain a first sequence; the first characteristic inverse function is an inverse function of the first timing characteristic function; at each node, inverse function inversion is adopted, and without loss of generality, only one sequence input can be assumed at each stage, because a plurality of inputs can be equivalently transformed into a large sequence combined by a plurality of sequences during processing.
In an alternative embodiment of the present invention, in step 124, outputting the original sequence of the time-series data by using the directed acyclic graph may include:
1241, randomly generating a plurality of groups of control parameter combinations according to a given key control parameter list of the judging rule; the control parameter list is the control parameter of the inverse function of the feature;
and 1242, selecting one group of control parameter combinations from the multiple groups of control parameter combinations, and generating corresponding original sequences by using the inverse function of the characteristics.
Specifically, in the counter-example generation module shown in fig. 3, the inversion task engine and the control parameter combination generation engine randomly generate a plurality of sets of control parameter combinations near the control parameter list, select a set of control parameter combinations, specifically, generate corresponding original quantities by using an inverse function, and randomly perturb other parts according to the change area setting.
In an alternative embodiment of the present invention, step 1242 may include:
step 12421, for expert rules with variables having rising or falling trends, generating a plurality of groups of expected index combination sets at the current value according to the size of the sliding window, the slope threshold and the proportion threshold;
step 12422, randomly generating a curve slope data set of a slope threshold value and a ratio threshold value combination value according with a rule according to the specified input array size m;
specifically, according to a ratio threshold, calculating the number n of the slope thresholds;
randomly generating n numerical values on the right side of the slope threshold, and randomly generating (m-n) numerical values on the left side of the slope threshold;
combining the numerical values into a slope array with the length of m according to a random sequence;
in step 12423, a corresponding original time sequence is generated according to the slope array with the length of m.
Optionally, generating a corresponding original time sequence according to the slope array with the length of m includes:
and in each time window, generating a standard straight line according to the slope in the slope array with the length of m, and then adding disturbance (polynomial disturbance, random noise and the like) to generate a corresponding original sequence.
Specifically, as shown in fig. 6, 7 and 8, example # 1, there are 3 control parameters: a sliding window size (windowSize), a slope threshold (sloputhreshold), a percentageThreshold (percentageThreshold);
generating 12 sets of expected index combinations at the current value, wherein the slope threshold is according to {2,2.5,3,3.5} and the proportion threshold is according to {0.2,0.3,0.4 };
randomly generating a curve slope data set (each combination generates 100 samples) which accords with the combination values of an expected slope threshold and a proportion threshold according to a specified input array size m (currently 10), and calculating the number n of the curve slope data set which is larger than the slope threshold according to the proportion threshold;
on the right side (near) of the slope threshold, n numerical values are randomly generated; randomly generating (m-n) numbers on the left side (near) of the slope threshold;
combining the numbers into an array with the length of m according to a random sequence;
according to the slope array, generating one number in each slope array of the corresponding original time sequence, and generating an array with the size of a sliding window: first, an array is generated from a standard straight line from the slope (which may also be generated from a polynomial) and then perturbed randomly.
For example, if the rule is "the temperature rises more than 5 degrees in 10 seconds with a ratio of more than 30%", assuming a temperature sensor sampling period of 1 second, and the rule runs once per minute, then the original sequence is "a time series of temperatures of length 60", and the slope array length is 6 (one slope is calculated every 10 seconds). First we generate a slope array, e.g. [5.1, -1, -2,6, -10,0], from a slope > 30% of >5 (i.e. [ 6 ═ 30%), and then generate a temperature sequence of length 10 (plus random perturbations) from each specific value in the slope array, e.g. [55,54,51,51,50,48,47,46,45,45], from the 2 nd value "-1" of the slope array.
In another alternative embodiment of the present invention, step 1242 may comprise:
step 12431, determining the position of the frequency in the FFT array according to the size of the sliding window and the rotating speed for the expert rule that the frequency corresponding to the rotating speed of the machine is in the ascending or descending trend;
step 12432, randomly generating a phase according to the amplitude of the frequency, and determining the value of the phase in the FFT array;
step 12433, generating a complex number randomly for the rest positions;
in step 12434, the FFT array is inverse FFT transformed to obtain the corresponding original sequence.
Specifically, as shown in fig. 9 and 10, a 1X amplitude change timing sequence is generated according to an example # 1 logic, and for one point of the 1X amplitude change timing sequence, the position of 1X in the FFT array is determined according to the window size and the rotation speed; randomly generating a phase according to the amplitude of 1X, and determining the numerical value of the phase in the FFT array; for the rest positions, randomly generating complex numbers; and performing inverse FFT on the FFT array to obtain a corresponding time sequence.
For example, assume the rule is "1 minute, 1 octave amplitude of vibration rises more than 5um in 10 seconds at a rate > 30%"; the vibration sensor acquisition frequency is assumed to be 1024 Hz. First, in a manner similar to example 1, a 1-time-multiplication amplitude sequence a of length 60 meeting the judgment condition is generated from the last-stage node, and this example illustrates how an original vibration quantity array of length 1024 is generated for a specific amplitude value in the sequence a.
In yet another alternative embodiment of the present invention, step 1242 may comprise:
step 12441, generating standard rolling curves from a plurality of historical rolling processes for the expert rules of gradual degradation of the times of rolling current wave head lacking;
12442, randomly selecting a plurality of wave crest areas according to the times of wave lack heads;
step 12443, generating a plurality of flat wave head fragments with different lengths near the wave crest according to the minimum length and the maximum change rate;
and step 12444, replacing the original wave head with the generated flat wave head to obtain the corresponding original sequence.
Specifically, as shown in fig. 11 and 12, according to the example # 1 logic, a variation time series of the missing header number is generated (except that the number type is an integer); generating a standard rolling curve from a plurality of historical rolling processes for each wave head missing event; randomly selecting a plurality of wave crest areas according to the times of wave lack heads; generating a plurality of flat wave head segments with different lengths near the wave crest according to the minimum length and the maximum change rate; and replacing the original wave head with the generated flat wave head.
In an optional embodiment of the present invention, the method for generating the counter example of the expert rule system based on the time series data may further include:
and step 13, performing cross processing between cases or variation processing of the cases on the generated original sequence, and outputting.
As shown in fig. 3, the counter-example generation module may further perform variation processing or cross processing on the input original sequence, as shown in fig. 13, where the variation is to add random disturbance noise to the original sequence, and the form and parameters of the noise allow the user to select; for example, when performing Case-to-Case variation processing on an original sequence (e.g., the boundary Case #1), the original sequence may be disturbed to obtain a varied sequence;
as shown in fig. 14, the crossing is a combination of a plurality of sequences generated at random, for example, a linear addition of the boundary Case # 1 and the boundary Case # 2, a random value extraction from the sequences, and the like, and after the crossing, a further perturbation may be performed.
In an optional embodiment of the present invention, the method for generating the counter example of the expert rule system based on the time series data may further include:
and 14, performing at least one of marginal evaluation, disturbance and case screening on the output original sequence to obtain a final output sequence.
In order for the generated original sequence to be more effective for rule verification, it is desirable that the generated sequence be at the edge of the rule being evaluated. That is, the sequence can randomly walk between satisfying and failing the rule by adding small random perturbations. The sequence generated by each inverse function is usually at the edge, but after many deductions and intersections/variations, the final partial original sequence may deviate from the edge, and therefore, a certain evaluation and screening mechanism is required.
Specifically, a random disturbance generator may be used to generate random disturbance noise. The evaluation indexes of the marginality can select purity indexes such as information entropy, Gini coefficient and the like, and the distribution proportion of the satisfied cases and the unsatisfied cases is checked. Case screening can select sequences that compare close edges based on a specified marginality threshold.
The original sequence generated by the counter-example generation module shown in fig. 3 or the sequence after the mutation or cross processing is subjected to at least one of the evaluation of the marginality, the disturbance and the case screening.
As shown in fig. 15, for the boundary case screening, several groups of noise are generated, and the case after the noise is added, if a positive/negative case ratio with a certain ratio can be ensured, the case to be evaluated is considered as the boundary case.
According to the embodiment of the invention, aiming at the early warning rule of time sequence data (such as common sensor data in industry), the method for generating the boundary case in an inversion mode is adopted according to the DAG of the rule, the boundary case is judged according to the boundary condition, a large number of boundary cases are generated through operations such as boundary case intersection, variation and the like, the generated cases can continuously approach the limit of expert experience, and the quantification and the precision of expert knowledge are accelerated.
An embodiment of the present invention further provides a counter example generating device of an expert rule system based on time series data, where the generating device may be an architecture as shown in fig. 3, and the counter example generating device includes:
the acquisition module is used for acquiring expert rules of an expert rule system based on the time sequence data of the industrial equipment;
and the processing module is used for adopting a counter example generation module according to the expert rule and outputting the original sequence of the time sequence data.
Optionally, according to the expert rule, outputting the original sequence of the time sequence data by using a counter-example generation module, includes:
inputting the expert rules into an oriented acyclic graph parser of the expert rules for parsing to obtain an oriented acyclic graph of the expert rules;
finding out all path sequences from a first-stage node to a last-stage node in the directed acyclic graph;
on the last-stage node of one of the channel sequences, inputting the transformed sequence output by the first time sequence characteristic function into a first characteristic inverse function corresponding to the node to obtain a first sequence; the first characteristic inverse function is an inverse function of the first timing characteristic function;
inputting the first sequence into a second characteristic inverse function corresponding to a previous node of a last node of the directed acyclic graph to obtain a second sequence until the directed acyclic graph outputs an original sequence of the time sequence data; the second characteristic inverse function is an inverse function of the second characteristic function corresponding to the previous-stage node.
Optionally, outputting the original sequence of the time series data by the directed acyclic graph includes:
randomly generating a plurality of groups of control parameter combinations according to a key control parameter list of a given judging rule; the control parameter list is the control parameter of the inverse function of the feature;
and selecting one group of control parameter combinations from the plurality of groups of control parameter combinations, and generating corresponding original sequences by utilizing the inverse function of the characteristics.
Optionally, selecting a group of control parameter combinations from the plurality of groups of control parameter combinations, and generating a corresponding original sequence by using a characteristic inverse function, including:
generating a plurality of groups of expected index combination sets at the current numerical value according to the size of the sliding window, the slope threshold and the proportion threshold;
randomly generating a curve slope data set of a slope threshold value and a ratio threshold value combination value which accord with a judging rule according to the size m of a specified input array;
and generating a corresponding original time sequence according to the curve slope data set.
Optionally, generating a corresponding original time sequence according to the slope array with the length of m includes:
and in each time window, generating a standard straight line according to the slope in the slope array with the length of m, and then generating a corresponding original sequence after scrambling.
Optionally, selecting a group of control parameter combinations from the plurality of groups of control parameter combinations, and generating a corresponding original sequence by using a characteristic inverse function, including:
determining the position of the frequency in the Fast Fourier Transform (FFT) array according to the size of the sliding window and the rotating speed;
randomly generating a phase according to the amplitude of the frequency, and determining the numerical value of the phase in the FFT array;
for the rest positions, randomly generating complex numbers;
and performing inverse FFT on the FFT array to obtain a corresponding original sequence.
Optionally, selecting a group of control parameter combinations from the plurality of groups of control parameter combinations, and generating a corresponding original sequence by using a characteristic inverse function, including:
generating a standard rolling curve from a plurality of historical rolling processes;
randomly selecting a plurality of wave crest areas according to the times of wave lack heads;
generating a plurality of flat wave head fragments with different lengths near the wave crest according to the minimum length and the maximum change rate;
and replacing the original wave head with the generated flat wave head to obtain a corresponding original sequence.
Optionally, the method for generating a counter example of the expert rule system based on the time series data further includes:
and performing cross processing among cases and/or variation processing of the cases on the generated original sequence, and outputting.
Optionally, the method for generating a counter example of the expert rule system based on the time series data further includes:
and performing at least one of marginal degree evaluation, disturbance and case screening on the output original sequence to obtain a final output sequence.
It should be noted that the apparatus is an apparatus corresponding to the embodiment of the method shown in fig. 1, and all the implementations in the embodiment of the method are applicable to the embodiment of the apparatus, so that the same technical effects can be achieved.
Furthermore, it is to be noted that in the device and method of the invention, it is obvious that the individual components or steps can be decomposed and/or recombined. These decompositions and/or recombinations are to be regarded as equivalents of the present invention. Also, the steps of performing the series of processes described above may naturally be performed chronologically in the order described, but need not necessarily be performed chronologically, and some steps may be performed in parallel or independently of each other. It will be understood by those skilled in the art that all or any of the steps or elements of the method and apparatus of the present invention may be implemented in any computing device (including processors, storage media, etc.) or network of computing devices, in hardware, firmware, software, or any combination thereof, which can be implemented by those skilled in the art using their basic programming skills after reading the description of the present invention.
Thus, the objects of the invention may also be achieved by running a program or a set of programs on any computing device. The computing device may be a general purpose device as is well known. The object of the invention is thus also achieved solely by providing a program product comprising program code for implementing the method or the apparatus. That is, such a program product also constitutes the present invention, and a storage medium storing such a program product also constitutes the present invention. It is to be understood that the storage medium may be any known storage medium or any storage medium developed in the future. It is further noted that in the apparatus and method of the present invention, it is apparent that each component or step can be decomposed and/or recombined. These decompositions and/or recombinations are to be regarded as equivalents of the present invention. Also, the steps of executing the series of processes described above may naturally be executed chronologically in the order described, but need not necessarily be executed chronologically. Some steps may be performed in parallel or independently of each other.
While the foregoing is directed to the preferred embodiment of the present invention, it will be understood by those skilled in the art that various changes and modifications may be made without departing from the spirit and scope of the invention as defined in the appended claims.
Claims (9)
1. A counter-example generation method of an expert rule system based on time series data is characterized by comprising the following steps:
acquiring an expert rule of an expert rule system based on time sequence data of industrial equipment;
according to the expert rule, a counter example generation module is adopted to output the original sequence of the time sequence data;
according to the expert rules, a counter-example generation module is adopted to output the original sequence of the time sequence data, and the method comprises the following steps:
inputting the expert rules into an oriented acyclic graph parser of the expert rules for parsing to obtain an oriented acyclic graph of the expert rules;
finding out all path sequences from a first-stage node to a last-stage node in the directed acyclic graph;
on the last-stage node of one of the channel sequences, inputting the transformed sequence output by the first time sequence characteristic function into a first characteristic inverse function corresponding to the node to obtain a first sequence; the first characteristic inverse function is an inverse function of the first timing characteristic function;
inputting the first sequence into a second characteristic inverse function corresponding to a previous node of a last node of the directed acyclic graph to obtain a second sequence until the directed acyclic graph outputs an original sequence of the time sequence data; the second characteristic inverse function is an inverse function of the second characteristic function corresponding to the previous-stage node.
2. The method of claim 1, wherein outputting the original sequence of time series data by a directed acyclic graph comprises:
randomly generating a plurality of groups of control parameter combinations according to a key control parameter list of a given judging rule; the control parameter list is the control parameter of the inverse function of the feature;
and selecting one group of control parameter combinations from the plurality of groups of control parameter combinations, and generating corresponding original sequences by utilizing the inverse function of the characteristics.
3. A method for generating an inverse example of an expert rule system based on time series data as claimed in claim 2, wherein selecting a group of control parameter combinations from a plurality of groups of control parameter combinations, and generating corresponding original sequences by using an inverse function of the features comprises:
generating a plurality of groups of expected index combination sets at the current numerical value according to the size of the sliding window, the slope threshold and the proportion threshold;
randomly generating a curve slope data set of a slope threshold value and a ratio threshold value combination value which accord with a judging rule according to the size m of a specified input array;
and generating a corresponding original sequence according to the curve slope data set.
4. A method for counter-example generation of expert rules system based on time series data according to claim 3, characterized in that generating corresponding original time series according to the slope array with length m comprises:
and in each time window, generating a standard straight line according to the slope in the slope array with the length of m, and then generating a corresponding original sequence after scrambling.
5. A method for generating an inverse example of an expert rule system based on time series data as claimed in claim 2, wherein selecting a group of control parameter combinations from a plurality of groups of control parameter combinations, and generating corresponding original sequences by using an inverse function of the features comprises:
determining the position of the frequency in the Fast Fourier Transform (FFT) array according to the size of the sliding window and the rotating speed;
randomly generating a phase according to the amplitude of the frequency, and determining the numerical value of the phase in the FFT array;
for the rest positions, randomly generating complex numbers;
and performing inverse FFT on the FFT array to obtain a corresponding original sequence.
6. A method for generating an inverse example of an expert rule system based on time series data as claimed in claim 2, wherein selecting a group of control parameter combinations from a plurality of groups of control parameter combinations, and generating corresponding original sequences by using an inverse function of the features comprises:
generating a standard rolling curve from a plurality of historical rolling processes;
randomly selecting a plurality of wave crest areas according to the times of wave lack heads;
generating a plurality of flat wave head fragments with different lengths near the wave crest according to the minimum length and the maximum change rate;
and replacing the original wave head with the generated flat wave head to obtain a corresponding original sequence.
7. A counter-example generation method of an expert rule system based on time series data according to claim 1, characterized by further comprising:
and performing cross processing among cases and/or variation processing of the cases on the generated original sequence, and outputting.
8. A counter-example generation method of an expert rule system based on time series data according to claim 1 or 7, characterized by further comprising:
and performing at least one of marginal degree evaluation, disturbance and case screening on the output original sequence to obtain a final output sequence.
9. An apparatus for generating counter examples of an expert rule system based on time series data, comprising: the acquisition module is used for acquiring expert rules of an expert rule system based on the time sequence data of the industrial equipment;
the processing module is used for generating a module by adopting a counter example according to the expert rule and outputting an original sequence of the time sequence data;
according to the expert rules, a counter-example generation module is adopted to output the original sequence of the time sequence data, and the method comprises the following steps:
inputting the expert rules into an oriented acyclic graph parser of the expert rules for parsing to obtain an oriented acyclic graph of the expert rules;
finding out all path sequences from a first-stage node to a last-stage node in the directed acyclic graph;
on the last-stage node of one of the channel sequences, inputting the transformed sequence output by the first time sequence characteristic function into a first characteristic inverse function corresponding to the node to obtain a first sequence; the first characteristic inverse function is an inverse function of the first timing characteristic function;
inputting the first sequence into a second characteristic inverse function corresponding to a previous node of a last node of the directed acyclic graph to obtain a second sequence until the directed acyclic graph outputs an original sequence of the time sequence data; the second characteristic inverse function is an inverse function of the second characteristic function corresponding to the previous-stage node.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010090753.1A CN111310113B (en) | 2020-02-13 | 2020-02-13 | Counter example generation method and device of expert rule system based on time sequence data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010090753.1A CN111310113B (en) | 2020-02-13 | 2020-02-13 | Counter example generation method and device of expert rule system based on time sequence data |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111310113A CN111310113A (en) | 2020-06-19 |
| CN111310113B true CN111310113B (en) | 2021-01-15 |
Family
ID=71145427
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010090753.1A Active CN111310113B (en) | 2020-02-13 | 2020-02-13 | Counter example generation method and device of expert rule system based on time sequence data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111310113B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111709528B (en) * | 2020-08-18 | 2021-01-05 | 北京工业大数据创新中心有限公司 | Expert rule protection method and device |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH08106391A (en) * | 1994-10-04 | 1996-04-23 | Fuji Electric Co Ltd | Discrimination rule generation method |
| US5644686A (en) * | 1994-04-29 | 1997-07-01 | International Business Machines Corporation | Expert system and method employing hierarchical knowledge base, and interactive multimedia/hypermedia applications |
| CN1547149A (en) * | 2003-12-04 | 2004-11-17 | 上海交通大学 | Method for implementing brain glioma computer aided diagnosis system based on data mining |
| CN101021842A (en) * | 2007-03-09 | 2007-08-22 | 清华大学 | Automatic learning and extending evolution handling method for Chinese basic block descriptive rule |
| CN102981841A (en) * | 2011-11-08 | 2013-03-20 | 微软公司 | External serialization and deserialization |
| CN103036730A (en) * | 2011-09-29 | 2013-04-10 | 西门子公司 | Method and device for achieving safety testing on protocol implementation |
| CN104346419A (en) * | 2013-07-25 | 2015-02-11 | 株式会社日立制作所 | Database analysis apparatus and method |
| CN104919445A (en) * | 2012-12-10 | 2015-09-16 | 起元技术有限责任公司 | System for transform generation |
| CN106201881A (en) * | 2016-07-12 | 2016-12-07 | 桂林电子科技大学 | A kind of CSP concurrent system adjustment method based on ASP |
| CN110710159A (en) * | 2017-05-31 | 2020-01-17 | 思科技术公司 | Generation of counter-examples for network intent form peer-to-peer failures |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102663191B (en) * | 2012-04-09 | 2014-07-23 | 西安电子科技大学 | SAT (satisfiability) based method for bounded model checking (BMC) for propositional projection temporal logic (PPTL) |
| CN105205112A (en) * | 2015-09-01 | 2015-12-30 | 西安交通大学 | System and method for excavating abnormal features of time series data |
-
2020
- 2020-02-13 CN CN202010090753.1A patent/CN111310113B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5644686A (en) * | 1994-04-29 | 1997-07-01 | International Business Machines Corporation | Expert system and method employing hierarchical knowledge base, and interactive multimedia/hypermedia applications |
| JPH08106391A (en) * | 1994-10-04 | 1996-04-23 | Fuji Electric Co Ltd | Discrimination rule generation method |
| CN1547149A (en) * | 2003-12-04 | 2004-11-17 | 上海交通大学 | Method for implementing brain glioma computer aided diagnosis system based on data mining |
| CN101021842A (en) * | 2007-03-09 | 2007-08-22 | 清华大学 | Automatic learning and extending evolution handling method for Chinese basic block descriptive rule |
| CN103036730A (en) * | 2011-09-29 | 2013-04-10 | 西门子公司 | Method and device for achieving safety testing on protocol implementation |
| CN102981841A (en) * | 2011-11-08 | 2013-03-20 | 微软公司 | External serialization and deserialization |
| CN104919445A (en) * | 2012-12-10 | 2015-09-16 | 起元技术有限责任公司 | System for transform generation |
| CN104346419A (en) * | 2013-07-25 | 2015-02-11 | 株式会社日立制作所 | Database analysis apparatus and method |
| CN106201881A (en) * | 2016-07-12 | 2016-12-07 | 桂林电子科技大学 | A kind of CSP concurrent system adjustment method based on ASP |
| CN110710159A (en) * | 2017-05-31 | 2020-01-17 | 思科技术公司 | Generation of counter-examples for network intent form peer-to-peer failures |
Non-Patent Citations (1)
| Title |
|---|
| 基于形式化规约的缺陷规则库构建与检测方法;佟超 等;《计算机工程与应用》;20141231;第50卷(第13期);第66-72、102页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111310113A (en) | 2020-06-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Yan et al. | Unsupervised learning for fault detection and diagnosis of air handling units | |
| Leisch et al. | Monitoring structural changes with the generalized fluctuation test | |
| Bagnall et al. | A run length transformation for discriminating between auto regressive time series | |
| CN113868006B (en) | Time sequence detection method and device, electronic equipment and computer storage medium | |
| CN111324637B (en) | Fault symptom searching method and system for industrial time sequence data | |
| CN111310113B (en) | Counter example generation method and device of expert rule system based on time sequence data | |
| CN111241673A (en) | Health state prediction method for industrial equipment in noisy environment | |
| CN108204341B (en) | method and device for identifying operating state of wind power plant | |
| CN107844731A (en) | Long-term sequence δ abnormal point detecting methods based on probabilistic suffix tree | |
| JP7119820B2 (en) | Prediction program, prediction method and learning device | |
| Elsner et al. | Complexity and predictability of hourly precipitation | |
| CN114861774A (en) | False data identification method and system in power grid | |
| Zhang et al. | Anomaly Detection of ICS based on EB-OCSVM | |
| CN108955855A (en) | Vibration signal characteristics extracting method, monitoring method and device towards rotating machinery | |
| TayebiHaghighi et al. | Fault diagnosis of rotating machine using an indirect observer and machine learning | |
| US11449781B2 (en) | Plant abnormality prediction system and method | |
| Aiordachioaie | On change detection and process diagnosis based on information and frequency domain models | |
| Scully et al. | Measuring chaos in the Lorenz and Rössler models: Fidelity tests for reservoir computing | |
| Noering et al. | Assessment and adaption of pattern discovery approaches for time series under the requirement of time warping | |
| Worden et al. | Damage identification using multivariate statistics: kernel discriminant analysis | |
| Aiordachioaie | Industrial process diagnosis based on information and time domain models | |
| Santos et al. | SED-bench: Benchmark for simulated electrical disturbance | |
| CN110717277A (en) | Time sequence wind speed simulation method | |
| Cao et al. | A Fast Randomized Algorithm for Finding the Maximal Common Subsequences | |
| CN116625685B (en) | Rolling bearing combination fault diagnosis method based on multi-domain feature construction |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |