Intelligent operation inspection method based on professional matching driving

Abstract

The invention provides an intelligent operation inspection method based on professional matching driving, and belongs to the technical field of intelligent operation inspection. Comprising the following steps: S1, generating multi-modal interaction content time sequence data based on an instruction, a reply and a final decision of an operation inspection person; based on the statistical instruction interaction time and interaction content similarity and the instruction complexity, the similarity of a strategy and an intelligent model is adopted to generate the similarity of the strategy; and calculating the professional matching degree difference of each interaction strategy according to the strategy finally adopted by the operation and maintenance personnel and the difference between the strategies in the intermediate interaction process. And for the same instruction, calculating the professional matching degree of the operation and maintenance strategy. S2, optimizing an operation inspection intelligent model based on a professional matching degree; comprising longitudinal training and transverse training. According to the method, the strategy matching performance under interaction of different instructions can be accurately evaluated, and the operation inspection efficiency and quality are effectively improved; the cross-individual learning and expansion of instruction connotation are realized, and the instruction understanding ability of the model is enriched in advance.

Description

Technical Field

[0001] The present invention provides an intelligent operation and maintenance method driven by professional matching, belonging to the technical field of intelligent operation and maintenance. Background Art

[0002] With the continuous expansion of the scale of the power system and the continuous improvement of the intelligent level, intelligent operation and maintenance technology plays an increasingly important role in power equipment condition monitoring, fault diagnosis, and operation and maintenance decision-making. Traditional operation and maintenance methods mainly rely on manual experience and rule-driven expert systems, making it difficult to cope with the challenges of a wide variety of power equipment, complex operating environments, and large amounts of data. In the scenario of multi-operator collaborative operation, there is a lack of a dynamic matching mechanism for operation and maintenance decision-making and instruction interaction of operators, which affects the overall operation and maintenance efficiency. Therefore, there is an urgent need for an intelligent operation and maintenance method driven by professional matching to achieve efficient coordination of power equipment condition monitoring, fault diagnosis, and operation and maintenance decision-making, thereby improving the intelligent operation and maintenance level of the power system.

[0003] Existing intelligent operation and maintenance methods cannot accurately calculate the professional matching degree of the operation and maintenance strategies given by the intelligent model in each interaction, making it difficult to accurately evaluate the strategy matching performance under different instruction interactions, and not considering the similarity of the interaction content of operation and maintenance personnel, making it difficult to achieve cross-individual learning and expansion of the instruction connotation. To address the above problems, the present invention proposes an intelligent operation and maintenance method driven by professional matching. First, a calculation method for the professional matching degree of operation and maintenance strategies based on interaction timing is designed, comprehensively considering instruction complexity, interaction time, interaction content similarity, and the similarity between the strategies actually adopted by operation and maintenance personnel and those generated by the intelligent model, to accurately calculate the professional matching degree of the operation and maintenance strategies given by the intelligent model in each interaction, thereby accurately evaluating the strategy matching performance under different instruction interactions. Second, an optimization method for the operation and maintenance intelligent model based on the professional matching degree is proposed, to improve the model performance through a horizontal and vertical hybrid training strategy, achieve cross-individual learning and expansion of the instruction connotation, and pre-enrich the instruction understanding ability of the model, thereby comprehensively optimizing the performance of the operation and maintenance intelligent model.

[0004] CN202411102224.3 discloses an intelligent operation and maintenance system for a thermal power plant based on big data and the Internet of Things. By collecting the load indexes of the working temperatures of boilers and steam turbines, calculating the efficiency indexes of energy input and output, and analyzing the environmental impact indexes of flue gas samples, the health status of the power generation system is judged by synthesizing the above three indexes, realizing real-time judgment of the safety of thermal power plant equipment. However, this technology does not comprehensively consider instruction complexity, interaction time, and interaction content similarity, cannot accurately calculate the professional matching degree of the operation and maintenance strategies given by the intelligent model in each interaction, and is difficult to accurately evaluate the strategy matching performance under different instruction interactions.

[0005] CN202311585431.4 discloses a power distribution area intelligent fusion terminal operation and maintenance system and an operation and maintenance method. An installation and operation and maintenance assistant is installed on a visual terminal, wirelessly connected to an operation and maintenance tool. Clicking on a detection / configuration item sends an instruction to the fusion terminal, and the terminal returns the result to the operation and maintenance assistant and displays it in real time on the visual terminal, realizing efficient and convenient remote operation and maintenance operations and management. However, this method cannot train interaction instructions in stages according to the professional matching degree sequence, making it difficult to deeply learn the potential instructions of operation and maintenance personnel, and it does not consider the similarity of the interaction content of operation and maintenance personnel, making it difficult to achieve cross-individual learning and expansion of the instruction connotation. Summary of the Invention

[0006] The present invention aims to solve the following technical problems:

[0007] 1. The problem that existing intelligent operation and maintenance methods do not comprehensively consider instruction complexity, interaction time, and interaction content similarity, and cannot accurately calculate the professional matching degree of the operation and maintenance strategies given by the intelligent model in each interaction.

[0008] 2. The problem that existing intelligent operation and maintenance methods do not consider the similarity of the interaction content of operation and maintenance personnel, making it difficult to achieve cross-individual learning and expansion of the instruction connotation.

[0009] The specific technical solution provided by the present invention is as follows:

[0010] 1. An intelligent operation and maintenance method driven by professional matching, characterized by including the following steps:

[0011] S1: Calculating the professional matching degree of operation and maintenance strategies based on the interaction time sequence;

[0012] First, based on the instructions, replies, and final decisions of operation and maintenance personnel, multi-modal interaction content time sequence data is generated.

[0013] Secondly, based on statistical instruction interaction time, interaction content similarity, and instruction complexity, the similarity between the strategy and the intelligent model for generating the strategy is adopted.

[0014] Then, according to the difference between the strategy finally adopted by the operation and maintenance personnel and the strategy during the intermediate interaction process, the professional matching degree difference of each interaction strategy can be calculated.

[0015] Finally, for the same instruction, calculate the professional matching degree of the operation and maintenance strategy.

[0016] S1.1 Generating multi-modal interaction content time sequence data;

[0017] To obtain multi-modal interaction content time sequence data, the system needs to capture instructions, replies, and decisions in real time. First, the system captures in real time the instructions issued by operation and maintenance personnel during the operation process. The instructions are in multi-modal form, that is, m(t) = (m ima (t), m voi(t), m text (t)).

[0018] Secondly, the system captures the responses of the operation and maintenance personnel to the system feedback; the response r(t) of the operation and maintenance personnel to the system feedback at the t-th time slot is decomposed into a multi-modal form, that is, r(t) = (r ima (t), r voi (t), r text (t)). Based on the captured instructions and responses, the system records the final decisions of the operation and maintenance personnel; based on the instructions, responses, and final decisions of the operation and maintenance personnel, multi-modal interaction content time-series data Data(t) is generated, expressed as:

[0019] Data(t) = {(m ima (t), m voi (t), m text (t)), r(t) = (r ima (t), r voi (t), r text (t)), n(t)}(1)

[0020] S1.2 Calculate the professional matching degree of the operation and maintenance strategy;

[0021] Statistical instruction interaction time, interaction content similarity, and instruction complexity, the similarity between the strategies adopted by the operation and maintenance personnel and the strategies generated by the intelligent model, calculate the professional matching of the operation and maintenance strategy;

[0022] First, by analyzing the interaction data between the instruction and the system, especially the similarity between the input instruction and the model response, to evaluate the system's mastery of the instruction. The higher the similarity, the better the mastery of the instruction, expressed as:

[0023]

[0024] Among them, S m (t) is the mastery degree of the input instruction m(t); sim(m(t), r model (t)) is the similarity between the input instruction m(t) and the model response r model (t). D m (t) = kδ m (t) is the interaction time of the instruction m(t) at the t-th time slot, δ m (t) is the instruction complexity of m(t), and k is an adjustment coefficient.

[0025] Secondly, based on the differences between the strategy finally adopted by the operation and maintenance personnel and the various strategies proposed during the intermediate interaction process, deeply analyze and calculate the difference in the professional matching degree of strategies at each interaction stage. Considering G interactions, calculate the difference in the professional matching degree P n (t) of each interaction strategy, which is expressed as:

[0026]

[0027] where n g (t) represents the strategy adopted in the g-th interaction, and n final (t) represents the strategy finally adopted by the operation and maintenance personnel, and G represents the total number of interactions.

[0028] Finally, calculate the professional matching degree Q m,n (t) of the operation and maintenance strategy based on the difference in the degree of instruction mastery and the professional matching degree of the interaction strategy, which is expressed as:

[0029]

[0030] S2: Optimization of the operation and maintenance intelligent model based on professional matching degree;

[0031] S2.1 Vertical training;

[0032] Implement a phased learning strategy by using the sequence of professional matching degrees during the interaction process of the same type of instructions for training.

[0033] First, conduct preliminary training for the interaction of instructions with low matching degrees; subsequently, conduct training for the interaction of instructions with high matching degrees to strengthen the learning effect. On this basis, further expand the training scope and conduct systematic training for different categories of instructions. This process is achieved by dynamically adjusting the training weights, which is expressed as:

[0034]

[0035] where ω m,n (t) represents the training weight of the decision n(t) corresponding to the input instruction m(t), τ represents the boundary of the training duration, t - 1 < t ≤ τ is the early stage of training, and τ ≤ t < t + 1 is the later stage of training.

[0036] S2.2 Horizontal training;

[0037] Considering the similarity of the interaction content of different operation and maintenance personnel, aggregate the models trained based on the professional matching degree sequences of different operation and maintenance personnel, so as to realize the reference and expansion of the potential instruction connotations of different operation and maintenance personnel, and thus realize the pre-learning of instruction connotations. For the same instruction, the similarity of the interaction content of different operation and maintenance personnel is expressed as:

[0038]

[0039] Aggregate the models trained based on the professional matching degree sequences of different operation and maintenance personnel, which is expressed as:

[0040]

[0041] Among them, Y m,n (t) is the model parameter obtained by the interactive training of the decision-making n(t) of the operation and maintenance personnel and the instruction m(t). is the aggregated model parameter. represents the parameter change function, and α n is the aggregation weight of the decisions of different operation and maintenance personnel.

[0042] The technical effects of the technical solution of the present invention are as follows:

[0043] 1. The present invention proposes a method for calculating the professional matching degree of operation and maintenance strategies based on interactive time series. This method generates multi-modal interactive content time series data based on the instructions, replies, and final decisions of operation and maintenance personnel. By comprehensively considering the instruction complexity, interaction time, similarity of interaction content, and the similarity between the actual strategies adopted by operation and maintenance personnel and the strategies generated by intelligent models, accurately calculate the professional matching degree of the operation and maintenance strategies given by the intelligent model in each interaction, so as to accurately evaluate the strategy matching performance under different instruction interactions, and effectively improve the operation and maintenance efficiency and quality.

[0044] 2. The present invention proposes an optimization method for operation and maintenance intelligent models based on professional matching degree, and improves the model performance through a horizontal and vertical hybrid training strategy. Vertically, for the same operation and maintenance personnel, according to the professional matching degree sequence in the interaction process of the same type of instructions, first intensively train the interaction instructions with lower matching degrees, and then optimize the interaction instructions with higher matching degrees, so as to deeply learn the potential instructions of the operation and maintenance personnel; Horizontally, aggregate the models trained based on the professional matching degree sequences of different operation and maintenance personnel, and consider the similarity of the interaction content of the operation and maintenance personnel during the aggregation process, realize cross-individual learning and expansion of the instruction connotations, and pre-enrich the instruction understanding ability of the model, so as to comprehensively optimize the performance of the operation and maintenance intelligent model. Specific implementation manners

[0045] The intelligent operation and maintenance method based on professional matching drive in this embodiment includes the following steps:

[0046] S1: Calculate the professional matching degree of operation and maintenance strategies based on interactive time series;

[0047] First, generate multi-modal interactive content time series data based on the instructions, replies, and final decisions of operation and maintenance personnel.

[0048] Secondly, based on the statistical interaction time of instructions, the similarity of interaction content, and the complexity of instructions, the similarity between the adopted strategy and the strategy generated by the intelligent model is used.

[0049] Then, according to the difference between the strategy finally adopted by the operation and maintenance personnel and the strategy in the middle interaction process, the professional matching degree difference of each interaction strategy can be calculated.

[0050] Finally, for the same instruction, calculate the professional matching degree of the operation and maintenance strategy. Divide the total optimization time slot into T equal-length time slots, and its set is represented as The length of each time slot is τ. At the t-th time slot, a total of M(t) instructions are considered, and the set is represented as M(t) = {1,..., m(t),..., M(t)}. Consider N(t) operation and maintenance personnel, and make a decision at each time slot. To simplify the variables, the operation and maintenance personnel and their corresponding decisions share the same set and index, and the set is represented as N(t) = {1,..., n(t),..., N(t)}.

[0051] S1.1 Generate multi-modal interaction content time-series data;

[0052] To obtain the multi-modal interaction content time-series data, the system needs to capture instructions, responses, and decisions in real time. First, the system needs to capture the instructions issued by the operation and maintenance personnel during the operation process in real time. The instructions are in multi-modal forms such as text, voice, or graphics, that is, m(t) = (m ima (t), m voi (t), m text (t)). Secondly, the system needs to capture the responses of the operation and maintenance personnel to the system feedback, including confirmation information, supplementary explanations, or corrected instructions, etc. The response r(t) of the operation and maintenance personnel to the system feedback at the t-th time slot can be decomposed into multi-modal forms such as text, voice, or graphics, that is, r(t) = (r ima (t), r voi (t), r text (t)). Based on the captured instructions and responses, the system needs to record the final decisions of the operation and maintenance personnel. This includes changes to system configurations, selection of fault handling solutions, adjustment of resource allocations, etc. The decision-making process may involve multiple steps and repeated confirmations. The system needs to accurately record the time points, contents, and context information of each decision. Based on the instructions, responses, and final decisions of the operation and maintenance personnel, generate the multi-modal interaction content time-series data Data(t), which is represented as:

[0053] Data(t) = {(m ima (t), m voi (t), m text (t)), r(t) = (r ima (t), r voi (t), r text(t)), n(t)}(1)

[0054] S1.2 Calculate the professional matching degree of operation and maintenance strategies;

[0055] Statistical instruction interaction time, interaction content similarity, and instruction complexity, the similarity between the strategies adopted by operation and maintenance personnel and the strategies generated by the intelligent model, calculate the professional matching of operation and maintenance strategies;

[0056] In order to comprehensively improve the efficiency of operation and maintenance work, statistical instruction interaction time, interaction content similarity, and instruction complexity are analyzed, and the similarity between the strategies adopted by operation and maintenance personnel in actual operations and the strategies automatically generated by the intelligent model is deeply analyzed. Considering the time efficiency of operation and maintenance personnel in responding to and handling problems, that is, the length of instruction interaction time, and measuring the matching degree between the instruction and the reply of the intelligent system, that is, the level of interaction content similarity. Furthermore, comparing the similarity between the actual operation strategies of operation and maintenance personnel and the strategies of the intelligent model, combining the above statistical and analysis results, calculating the professional matching degree of operation and maintenance strategies, promoting the standardization and intelligence of operation and maintenance work, and also providing data support and decision-making basis for continuously optimizing operation and maintenance processes and improving operation and maintenance efficiency.

[0057] First, by analyzing the interaction data between the instruction and the system, especially the similarity between the input instruction and the model reply, to evaluate the system's mastery of the instruction. The higher the similarity, the better the mastery of the instruction, expressed as:

[0058]

[0059] Among them, S m (t) is the mastery degree of the input instruction m(t); sim(m(t), r model (t)) is the similarity between the input instruction m(t) and the model reply r model (t). D m (t) = kδ m (t) is the interaction time of the instruction m(t) in the t-th time slot, δ m (t) is the instruction complexity of m(t), and k is the adjustment coefficient.

[0060] Secondly, according to the differences between the strategies finally adopted by operation and maintenance personnel and the various strategies proposed during the intermediate interaction process, deeply analyze and calculate the differences in the professional matching degree of strategies at each interaction stage. Not only focus on the final form of operation and maintenance personnel's decisions, but also pay more attention to the evolution and adjustment during the decision-making process, so as to more comprehensively evaluate the operation and maintenance decision-making ability. Considering G interactions, according to the differences between the strategies finally adopted by operation and maintenance personnel and the strategies during the intermediate interaction process, the difference in the professional matching degree P n (t) of each interaction strategy can be calculated, expressed as:

[0061]

[0062] Among them, n g (t) represents the strategy adopted in the g-th interaction, and n final (t) represents the strategy finally adopted by the operation and maintenance personnel, and G represents the total number of interactions.

[0063] Finally, calculate the professional matching degree Q of the operation and maintenance strategy based on the difference between the instruction mastery degree and the professional matching degree of the interaction strategy m,n (t), which is expressed as:

[0064]

[0065] S2: Optimization of the operation and maintenance intelligent model based on professional matching degree;

[0066] S2.1 Vertical training;

[0067] By using the professional matching degree sequence in the interaction process of the same type of instructions for training, implement a phased learning strategy. First, conduct preliminary training for the instruction interaction with a lower matching degree; subsequently, conduct training for the instruction interaction with a higher matching degree to strengthen the learning effect. On this basis, further expand the training scope and conduct systematic training for different categories of instructions. This process is achieved by dynamically adjusting the training weights, which is expressed as:

[0068]

[0069] Among them, ω m,n (t) represents the training weight of the decision n(t) corresponding to the input instruction m(t), τ represents the training duration boundary, t - 1 < t ≤ τ is the early stage of training, and τ ≤ t < t + 1 is the later stage of training.

[0070] S2.2 Horizontal training;

[0071] Considering the similarity of the interaction content of different operation and maintenance personnel, aggregate the models trained based on the professional matching degree sequences of different operation and maintenance personnel, so as to realize the reference and expansion of the potential instruction connotations of different operation and maintenance personnel, and thus realize the pre-learning of instruction connotations. For the same instruction, the similarity of the interaction content of different operation and maintenance personnel is expressed as:

[0072]

[0073] Aggregate the models trained based on the professional matching degree sequences of different operation and maintenance personnel, which is expressed as:

[0074]

[0075] Among them, Y m,n(t) is the model parameter for the interactive training of the operation and maintenance personnel's decision n(t) and the instruction m(t). is the aggregated model parameter. represents the parameter change function, α n is the aggregation weight for different operation and maintenance personnel's decisions.

Claims

1. An intelligent operation and maintenance method driven by professional matching, characterized in that Including the following steps: S1: Calculate the professional matching degree of operation and maintenance strategies based on the interaction time series; First, generate multi-modal interaction content time series data based on the instructions, replies, and final decisions of operation and maintenance personnel; Secondly, based on the statistical instruction interaction time, interaction content similarity, and instruction complexity, adopt the similarity between the strategy and the intelligent model to generate the strategy; According to the differences between the strategies finally adopted by the operation and maintenance personnel and the strategies in the intermediate interaction process, calculate the professional matching degree differences of each interaction strategy; finally, calculate the professional matching degree of the operation and maintenance strategy for the same instruction; S2: Optimize the operation and maintenance intelligent model based on the professional matching degree; S2.1 Vertical training; Implement a phased learning strategy by using the professional matching degree sequence in the interaction process of the same type of instructions for training; S2.2 Horizontal training; Consider the similarity of the interaction content of different operation and maintenance personnel, and aggregate the models trained based on the professional matching degree sequences of different operation and maintenance personnel, so as to realize the reference and expansion of the potential instruction connotations of different operation and maintenance personnel, and thus realize the pre-learning of instruction connotations.

2. The intelligent operation and inspection method based on professional matching drive according to claim 1, wherein S1 specifically includes the following sub-steps: S1.1 Generate multi-modal interaction content time series data; To obtain the time-series data of multi-modal interaction content, the system needs to capture instructions, responses, and decisions in real time. First, the system captures in real time the instructions issued by the operation and maintenance personnel during the operation process. The instructions are in multi-modal form, that is, m(t) = (m ima (t), m voi (t), m text (t)); Secondly, the system captures the responses of the operation and maintenance personnel to the system feedback; the response r(t) of the operation and maintenance personnel to the system feedback at the t-th time slot is decomposed into a multi-modal form, that is, r(t) = (r ima (t), r voi (t), r text (t)); based on the captured instructions and responses, the system records the final decisions of the operation and maintenance personnel; based on the instructions, responses, and final decisions of the operation and maintenance personnel, multi-modal interaction content time series data Data(t) is generated, expressed as: Data(t)={(m ima (t),m voi (t),m text (t)),r(t)=(r ima (t),r voi (t),r text (t)),n(t)} (1) S1.2 Calculate the professional matching degree of operation and maintenance strategies; Statistical instruction interaction time, interaction content similarity, and instruction complexity, the similarity between the strategies adopted by operation and maintenance personnel and the strategies generated by the intelligent model, calculate the professional matching of operation and maintenance strategies; First, evaluate the system's mastery of instructions by analyzing the interaction data between the instructions and the system, especially the similarity between the input instructions and the model replies; the higher the similarity, the better the mastery of the instructions, which is expressed as: Among them, S m (t) is the degree of mastery of the input instruction m(t); sim(m(t), r model (t)) is the similarity between the input instruction m(t) and the model response r model (t); D m (t) = kδ m (t) is the interaction time of the instruction m(t) in the t-th time slot, δ m (t) is the instruction complexity of m(t), and k is an adjustment coefficient; Secondly, based on the differences between the strategy finally adopted by the operation and maintenance personnel and the various strategies proposed during the intermediate interaction process, deeply analyze and calculate the professional matching degree differences of the strategies in each interaction stage; considering G interactions, calculate the professional matching degree difference P n (t) of each interaction strategy, which is expressed as: Among them, n g (t) represents the strategy adopted in the g-th interaction, n final (t) represents the strategy finally adopted by the operation and maintenance personnel, and G represents the total number of interactions; Finally, calculate the professional matching degree Q of the operation and maintenance strategy based on the difference in the professional matching degree between the instruction mastery level and the interaction strategy m,n (t), which is expressed as:

3. The intelligent operation and inspection method based on professional matching drive according to claim 1, wherein S2.1 Vertical training, the specific method is: First, conduct preliminary training for instruction interactions with low matching degrees; Subsequently, conduct training for instruction interactions with high matching degrees to strengthen the learning effect; On this basis, further expand the training scope and conduct systematic training for different categories of instructions; this process is achieved by dynamically adjusting the training weights, which is expressed as: Among them, ω m,n (t) represents the training weight of the decision n(t) corresponding to the input instruction m(t), τ represents the boundary of the training duration, t - 1 < t ≤ τ is the early stage of training, and τ ≤ t < t + 1 is the late stage of training.

4. The intelligent operation and inspection method based on professional matching drive according to claim 3, wherein In the horizontal training of S2.2, for the same instruction, the similarity of the interaction content among different operation and maintenance personnel It is expressed as: Aggregate the models trained based on the professional matching degree sequences of different operation and maintenance personnel, which is expressed as: Among them, Y m,n (t) is the model parameter for the interactive training of the operation and inspection personnel's decision-making n(t) and the instruction m(t). is the aggregated model parameter. represents the parameter change function, and α n is the aggregation weight of the decisions of different operation and inspection personnel.

Images