New energy centralized post evaluation management method and system based on large model

Abstract

The application discloses a new energy centralized control post evaluation management method and system based on a large model, relates to the technical field of evaluation management, and comprises the following steps: analyzing a new energy operation technical document, and constructing an operation semantic unit set; according to risk marking, giving risk weights and propagation coefficients to operation semantic paths, and forming a risk propagation structure; after reading an evaluation request, constructing a target operation semantic path, and establishing a structured evaluation semantic; inputting the structured evaluation semantic into a large model, and outputting evaluation questions; after obtaining an answering path corresponding to the evaluation questions, outputting a capability evaluation result and improvement suggestions. The application solves the technical problems of the prior art, such as dependence on manual paper setting and reading, strong subjectivity of evaluation, poor question setting pertinence, and low evaluation efficiency, achieves intelligent automatic question setting and intelligent evaluation scoring for new energy centralized control posts, and improves the technical effects of accuracy and intelligent efficiency of post evaluation.

Description

Technical Field

[0001] This invention relates to the field of evaluation and management technology, specifically to a method and system for evaluating and managing new energy centralized control positions based on a large model. Background Technology

[0002] Currently, many new energy centralized control positions rely on manual compilation of technical data, manual creation of assessment questions, and manual grading. There is a lack of standardized knowledge modeling and semantic question generation mechanisms, making it difficult to create targeted questions based on the actual operation process, risk relationships, and individual knowledge gaps of the personnel. At the same time, traditional assessment methods are highly subjective and arbitrary, have poor question adaptability, and inconsistent assessment standards. They cannot quantify the semantic relationships and risk propagation characteristics of job operations, nor can they automatically generate personalized assessment questions and intelligently conduct quantitative ability assessments and precise improvement guidance. This results in industry pain points such as low test paper generation efficiency, insufficient assessment accuracy, and a lack of scientific basis for job ability diagnosis.

[0003] Existing technologies suffer from problems such as reliance on manual test paper compilation and grading, high subjectivity in assessment, poor question targeting, and low assessment efficiency. Summary of the Invention

[0004] This application provides a method and system for evaluating and managing new energy centralized control positions based on a large model, which is used to address the technical problems of existing technologies that rely on manual test paper compilation and grading, have strong subjectivity in evaluation, poor question targeting, and low evaluation efficiency.

[0005] In view of the above problems, the first aspect of this application provides a method for evaluating and managing new energy centralized control positions based on a large model, the method comprising: The technical documents for new energy operation are parsed to extract information on the operation objects, operation actions, and state constraints. The extracted results are then transformed into quantifiable operation state feature vectors, and a set of operation semantic units containing risk labels and dependencies is constructed. Based on the temporal and dependency relationships of the set of operation semantic units, an operation semantic path is constructed. Risk weights and propagation coefficients are assigned to the operation semantic path according to the risk labels to form a risk propagation structure. After reading the assessment request, a target operation semantic path is constructed by matching the operation semantic path. After performing perturbation processing on the target operation semantic path, a structured assessment semantic is established. The structured assessment semantic is input into a large model to output assessment questions. After obtaining the answer path corresponding to the assessment questions, a weighted score is performed using the operation semantic path and the risk propagation structure to output the capability assessment results and improvement suggestions.

[0006] A second aspect of this application provides a new energy centralized control job assessment and management system based on a large model, the system comprising: The document parsing module parses technical documents related to new energy operation, extracts information on operational objects, actions, and state constraints, and transforms the extracted results into quantifiable operational state feature vectors. It then constructs a set of operational semantic units containing risk labels and dependencies. The risk propagation structure generation module generates operational semantic paths based on the temporal and dependency relationships of the operational semantic unit set. It assigns risk weights and propagation coefficients to these paths according to risk labels, forming a risk propagation structure. The structured assessment semantic establishment module, after reading the assessment request, matches and constructs a target operational semantic path from the operational semantic path, performs perturbation processing on the target operational semantic path, and establishes structured assessment semantics. The structured assessment semantic output module inputs the structured assessment semantics into a large model and outputs assessment questions. The weighted scoring module, after obtaining the answer path corresponding to the assessment questions, performs weighted scoring using the operational semantic path and risk propagation structure, outputting capability assessment results and improvement suggestions.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: This method involves parsing technical documents related to new energy operation and constructing a set of operational semantic units (ISUs) including risk labeling and dependencies. It then constructs operational semantic paths, assigning risk weights and propagation coefficients to these paths based on risk labeling to form a risk propagation structure. Upon reading an assessment request, it matches and constructs a target operational semantic path from these existing SMUs, performs perturbation processing on the target path, and establishes structured assessment semantics. This structured assessment semantics is input into a large model, outputting assessment questions. After obtaining the answer paths corresponding to the assessment questions, it utilizes the operational semantic paths and risk propagation structure to perform weighted scoring, outputting capability assessment results and improvement suggestions. This approach achieves the technical effect of intelligently and automatically generating questions and intelligently scoring assessments for new energy centralized control positions, improving the accuracy and efficiency of job assessments. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 A schematic diagram of the evaluation and management method for new energy centralized control positions based on a large model, provided in this application embodiment; Figure 2 This is a schematic diagram of the structure of the new energy centralized control post evaluation and management system based on a large model, provided in an embodiment of this application.

[0010] Figure labeling: Document parsing module 10, risk communication structure generation module 20, structured assessment semantic establishment module 30, structured assessment semantic output module 40, weighted scoring module 50. Detailed Implementation

[0011] This application provides a new energy centralized control post assessment management method and system based on a large model, which is used to address the technical problems of existing technologies that rely on manual test paper compilation and grading, have strong subjectivity in assessment, poor question targeting, and low assessment efficiency.

[0012] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0013] Example 1, as Figure 1 As shown, this application provides a method for evaluating and managing new energy centralized control positions based on a large model, the method including: Step S100: Parse the technical documents for new energy operation, extract the operation objects, operation actions and state constraint information, and transform the extraction results into quantifiable operation state feature vectors to construct a set of operation semantic units that include risk labeling and dependency relationships.

[0014] Specifically, the operation technical documents such as operation specifications and maintenance manuals in the new energy centralized control scenario are structured and parsed to extract three core elements: operation objects, operation actions, and state constraint information. The unstructured text information is then converted into numerical and vectorized forms to form operation state feature vectors that can be directly computed by the model. Based on the extraction results, the risk level of each smallest semantic unit is labeled and the execution dependency relationship between units is established. Finally, a set of operation semantic units containing risk labels and dependency relationships is constructed.

[0015] Step S200: Based on the temporal and dependency relationships of the set of runtime semantic units, construct runtime semantic paths, assign risk weights and propagation coefficients to the runtime semantic paths according to risk labeling, and form a risk propagation structure.

[0016] Specifically, based on the inherent temporal relationships and execution logic dependencies among the units in the constructed set of operational semantic units, discrete semantic units are sequentially connected according to the actual process of new energy centralized control operation to form a complete and coherent operational semantic path. According to the risk level pre-labeled for each semantic unit, each node in the operational semantic path is assigned a corresponding risk weight, and a propagation coefficient is allocated according to the dependence strength and transmission influence between adjacent units. Finally, a risk propagation structure that can fully characterize the transmission, amplification and attenuation of risk along the operation process is constructed.

[0017] Step S300: After reading the evaluation request, construct the target running semantic path from the running semantic path matching, and after performing perturbation processing on the target running semantic path, establish the structured evaluation semantics.

[0018] Specifically, the system reads and structurally parses externally input assessment requests, extracts job type tags, operational scenario tags, and capability dimension constraint information, and constructs a semantic constraint vector for precise matching. Based on this semantic constraint vector, it performs consistency scoring and fit calculation on all constructed operational semantic paths, sorts and selects target operational semantic paths that meet the assessment requirements. According to the dependencies and risk propagation characteristics within the target operational semantic paths, it selects key semantic units with high risk weights and dependency centrality, and performs perturbation processing such as deletion of operation information, offsetting of state constraints, or reconstruction of risk propagation relationships. The perturbed operational semantic paths are then standardized and organized to finally form a structured assessment semantic suitable for generating large-scale model test questions.

[0019] Step S400: Input the structured assessment semantics into the large model and output the assessment questions.

[0020] Specifically, firstly, the operational technical documents, operating procedures, and maintenance manuals in the field of new energy centralized control are uniformly analyzed, segmented, and vectorized to construct an assessment knowledge base covering operational objects, action logic, state constraints, and risk points. Based on this, a dedicated assessment model is built. The model adopts a three-level structure: a base model, a domain adaptation layer, and a constraint reasoning layer. The base model is used for general semantic understanding, the domain adaptation layer is used to integrate new energy centralized control professional knowledge, and the constraint reasoning layer is used to ensure the compliance and risk relevance of question generation. During training, semantic units and semantic paths are used for execution. Using the risk propagation structure as labeled samples, the model is trained through supervised fine-tuning, knowledge distillation, and comparative learning, enabling it to generate test questions, match knowledge points, and locate weaknesses. The large model is trained by inputting the perturbation-processed structured assessment semantics. Based on the job type, scenario constraints, and difficulty requirements in the assessment request, the model retrieves relevant content from the assessment knowledge base and generates assessment questions. When no assessment knowledge points are specified, it further retrieves the target user's historical test data, automatically analyzes frequently missed questions and weak knowledge points, and completes targeted test paper generation, ultimately outputting standardized assessment questions that meet the assessment requirements.

[0021] Step S500: After obtaining the answer path corresponding to the assessment question, perform weighted scoring using the semantic path and risk propagation structure, and output the ability assessment results and improvement suggestions.

[0022] Specifically, after obtaining the user's answer path for the assessment questions, the answer path is aligned with the standard operating semantic path in terms of time sequence and dependency. Abnormal nodes are marked and the state deviation is calculated. The deviation results are then mapped to the risk propagation structure. Risk accumulation and propagation analysis is performed along the propagation coefficient and edge propagation sensitivity to construct a risk deviation index. Weighted scoring is performed based on three dimensions: completeness of operation steps, accuracy of state judgment, and consistency of risk response. Simultaneously, multi-dimensional ability analysis is carried out on the answer data, and finally, objective and quantitative ability assessment results are output. Targeted improvement suggestions are generated based on weak points.

[0023] In one possible implementation, step S500 further includes: Step S510: Align the answer path with the running semantic path, locate the valid matching sub-path in the answer path based on the sequential consistency of semantic units and the degree of dependency matching, and mark the unmatched semantic units as abnormal nodes.

[0024] Step S520: Use the running state feature vectors corresponding to each semantic unit to perform a consistency comparison of the answer path and establish the state deviation quantity.

[0025] Step S530: After mapping the abnormal nodes and state deviations to the risk propagation structure, perform risk accumulation propagation analysis along the propagation coefficients of the semantic path to establish a risk deviation index.

[0026] Step S540: Obtain the coverage and risk deviation index of the effective matching sub-paths, and calculate the operation step integrity score, status judgment accuracy score, and risk response consistency score to complete the weighted scoring.

[0027] Specifically, the system performs bidirectional alignment of the user-generated answer path for the assessment questions with the system's preset standard operating semantic path, considering both temporal order and logical dependencies. Based on the consistency of the semantic units in the process and the degree of matching of dependencies between units, it conducts a node-by-node comparison, accurately identifies and extracts valid matching sub-paths that are completely consistent with the standard path from the answer path, and marks any missing, out-of-order, dependency-violated, or non-corresponding content in the answer path as abnormal nodes.

[0028] For the semantic units that are effectively matched, the corresponding running state feature vectors under the user's answer path and the standard running semantic path are extracted respectively. The two sets of vectors are first normalized, and then the similarity between the vectors is calculated by the cosine similarity formula. At the same time, the absolute difference of the feature values ​​is calculated dimension by dimension using the Euclidean distance formula. The similarity score and the dimensional difference are weighted and fused to finally generate a state deviation quantity that quantitatively reflects the difference between the user's answer and the standard state. The larger the deviation quantity, the more significant the deviation between the answer state and the standard state.

[0029] The identified abnormal nodes and their quantified state deviations are uniformly projected onto the corresponding semantic unit nodes within the risk propagation structure. Combining the specific classification type of the abnormal node with the magnitude of the state deviation, the inherent risk weights of all semantic unit nodes are dynamically reconstructed and updated one by one. This generates a dynamic risk propagation substructure that simultaneously includes single-node risk increments and inter-node edge propagation sensitivity. This edge propagation sensitivity is modeled and constructed based on the strength of the logical dependency between adjacent semantic units and the degree of chain reaction effect of the state deviation of the preceding unit on the following unit. Based on the constructed dynamic risk propagation substructure, using the reconstructed node risk increments as the initial risk propagation source, a layered and orderly risk propagation process is carried out strictly along the pre-set propagation coefficients and defined edge propagation sensitivity of the operational semantic path. The risk propagation iterative calculation continuously accumulates and integrates the transmission risk data transmitted by upstream neighboring nodes during the propagation process. Combined with its own bound operational status constraint information, it performs adaptive regulation processing in real time on the incoming propagation risks, such as absorption and weakening, amplification and strengthening, or truncation and blocking. Finally, it solves for the actual post-propagation risk value of each semantic unit node after transmission regulation. Based on the post-propagation risk value sequence of each node in the entire path, it continuously extracts the risk accumulation gradient change law and the characteristics of local risk value surge sections along the complete operational semantic path. It integrates multi-dimensional feature parameters such as deviation diffusion range, risk accumulation magnitude and key risk impact sections, and constructs a quantitative risk deviation index that can comprehensively characterize the diffusion evolution trend, harm range and key weak links of the answer deviation in the new energy operation process.

[0030] The system obtains the coverage of effective matching sub-paths and the generated risk deviation index. Coverage is calculated by dividing the number of semantic units contained in the effective matching sub-path by the total number of semantic units in the standard operating semantic path, i.e., Coverage = Number of effective matching semantic units / Total number of standard semantic units × 100%. Subsequently, three core scores are calculated. The operation step completeness score is based on coverage and uses a linear mapping formula: Completeness Score = Coverage × 100. If key semantic units are missing, points are deducted based on the number of missing units multiplied by preset deduction items to ensure a strong correlation between the score and step completeness. The state judgment accuracy score is calculated based on the state deviation of each effective matching semantic unit. First, the state deviation of each semantic unit is normalized: Normalized Deviation = Actual Deviation / Maximum Allowable Deviation. Then, the accuracy score is calculated using the formula: Accuracy Score = 100 - Σ(Normalized Deviation × Corresponding Semantic Unit Weight). The smaller the deviation, the higher the score, accurately reflecting the user's judgment accuracy of each operating state. The risk response consistency score is calculated by reverse conversion based on the risk deviation index using an exponential decay formula. The risk deviation index is used to control the rate of score decay. The smaller the risk deviation index, the closer the score is to the full score, reflecting the degree of consistency between the user's answer and the standard risk response logic. Finally, the weights of the three scores are preset, such as 40% for completeness of operation steps, 30% for accuracy of status judgment, and 30% for consistency of risk response. Through the weighted summation formula, the comprehensive score = completeness score × 40% + accuracy score × 30% + consistency score × 30% is completed to complete the overall weighted score and output a quantitative comprehensive evaluation value.

[0031] In one possible implementation, step S530 further includes: Step S531: Project the abnormal nodes and state deviations onto the corresponding semantic unit nodes in the risk propagation structure. Reconstruct the risk weights of each semantic unit node based on the type of the abnormal node and the state deviation, and generate a dynamic risk propagation substructure containing node risk increments and edge propagation sensitivity. The edge propagation sensitivity is constructed based on the strength of the dependency relationship between adjacent semantic units and the degree of influence of state deviations on subsequent semantic units.

[0032] Step S532: In the dynamic risk propagation substructure, the reconstructed node risk increment is used as the propagation source. The propagation coefficient along the semantic path and the edge propagation sensitivity are used to perform hierarchical propagation calculation. During the propagation process, the propagation risk received from the upstream node is accumulated for each semantic unit node. The propagation risk is absorbed, amplified or truncated according to the state constraint information of the corresponding node to obtain the post-propagation risk value of each semantic unit node.

[0033] Step S533: Based on the post-propagation risk value of each semantic unit node, extract the risk accumulation gradient change characteristics and local sudden increase segments along the running semantic path, and construct a risk deviation index that reflects the diffusion trend of answer deviation in the running process and key impact segments based on the extraction results.

[0034] Specifically, a node mapping algorithm is used to accurately project identified and marked abnormal nodes and calculated state deviations onto semantic unit nodes in a pre-defined risk propagation structure that perfectly correspond to the semantic attributes and functional logic of the abnormal nodes, ensuring unbiased and one-to-one mapping. Based on the specific types of abnormal nodes, such as missing steps, incorrect sequences, and misjudged states, different type weight coefficients are preset. Combined with the normalized state deviation value, the normalization formula is: Normalized Deviation = Actual State Deviation / Maximum Allowable Deviation. Using a weighted summation formula, the node risk increment = Abnormal Node Type Weight × Normalized Deviation + Inherent Risk Value of the Semantic Unit. The original risk weights of each semantic unit node are then reconstructed and updated to obtain the node risk increment corresponding to each semantic unit node. Simultaneously, the Pearson correlation coefficient is used to calculate the strength of logical dependencies between adjacent semantic units. This is combined with the impact of the state deviation of the preceding semantic unit on the operational process, state judgment, and risk level of the following semantic unit. The impact is quantified using an impact coefficient obtained through training with historical operational data. A linear fusion formula is used: edge propagation sensitivity = 0.6 × dependency strength + 0.4 × deviation impact coefficient. The weights can be adjusted according to the actual scenario to construct the edge propagation sensitivity of each adjacent semantic unit's connecting edge. Finally, the node risk increments after all reconstructions and the edge propagation sensitivity of each connecting edge are integrated to generate a dynamic risk propagation substructure that can be used for subsequent risk propagation analysis. This substructure can adapt to changes in abnormal nodes and state deviation in real time, ensuring the accuracy of risk propagation analysis.

[0035] Within the constructed dynamic risk propagation substructure, the node risk increments calculated from the reconstruction of each semantic unit node serve as the initial source of risk propagation. Relying on a graph traversal iterative algorithm, and strictly following the flow order of the standard semantic path, combined with the path's preset fixed propagation coefficient and the edge propagation sensitivity quantified between adjacent nodes, a layered, progressive risk propagation iterative calculation is carried out from upstream to downstream. During the calculation, for each semantic unit node, the risk values ​​transmitted from all upstream associated nodes are accumulated and summarized in real time to form the total risk base received by the node. Simultaneously, the state constraint information bound to the current semantic unit node is retrieved. Through preset constraint judgment threshold rules, differentiated control processing is performed on the inflowing risks. When the state constraint has risk suppression attributes, an attenuation coefficient is used to absorb and weaken the risk; when the state constraint has a chain amplification risk, a gain coefficient is used to amplify and strengthen the risk; when the state constraint has isolation and blocking conditions, a risk truncation and zeroing operation is directly performed. After a full-link iterative calculation involving node-by-node risk superposition, cross-edge propagation conversion, and adaptive state constraint control, the post-propagation risk value of each semantic unit node after propagation correction is solved and output, providing accurate quantitative data support for subsequent risk feature extraction and indicator construction.

[0036] The system sequentially retrieves the post-propagation risk values ​​obtained from all semantic unit nodes after risk transmission and constraint control. A continuous risk value sequence is constructed according to the standard temporal arrangement of the operational semantic path. First-order difference operations are used to calculate the risk difference between adjacent semantic unit nodes point by point, obtaining the risk accumulation gradient for the entire path. This quantifies the continuous rate of change and accumulation trend of risk along the process. Simultaneously, a dynamic threshold discrimination mechanism is set, using the global risk mean and standard deviation as a benchmark to define a sudden increase threshold. The system traverses the risk value sequence and filters out locations where risk values ​​exceed the threshold range or where the increase exceeds a preset proportion within a short interval, accurately identifying and marking local risk surge segments. Subsequently, multiple quantitative parameters are integrated, including the risk accumulation gradient mean, gradient fluctuation amplitude, number of surge segments, surge peak size, and deviation diffusion coverage. A multi-factor weighted fusion algorithm is used to assign specific weights to each dimension of feature parameters and complete normalized fusion calculations. This uniformly quantifies the diffusion range, evolution trend, severity, and key dangerous impact segments of the answer deviation in the complete operational process. Finally, a standardized and quantifiable risk deviation index is generated, intuitively representing the cascading risk hazard level and process impact characteristics caused by answer errors.

[0037] In one possible implementation, step S533 further includes: Arrange the semantic unit nodes in the execution semantic path according to the execution order to construct the corresponding post-propagation risk value sequence.

[0038] The risk increment sequence between nodes is calculated based on the risk value after propagation of adjacent semantic unit nodes. The risk increment sequence is then subjected to constraint normalization processing using state constraint information to obtain a standardized risk increment that reflects the degree of risk evolution under changes in unit operating conditions.

[0039] Sliding cumulative calculation is performed on the standardized risk increment along the semantic path of operation to construct the risk accumulation change curve within the continuous semantic unit window. During the sliding process, continuous node intervals that satisfy the condition that the cumulative increment is continuously positive and the growth rate is monotonically increasing are identified and marked as risk accumulation segments.

[0040] At the end node of the risk accumulation section, the deviation between the actual risk value after propagation and the predicted risk value obtained by extrapolating the risk from the preceding node is calculated.

[0041] If the deviation exceeds the preset deviation threshold, the corresponding node will be identified as a risk release trigger node, and the risk accumulation section will be traced back from the corresponding node to determine the local surge section.

[0042] Specifically, in the pre-execution stage, the post-propagation risk value data of all semantic unit nodes within the risk propagation substructure, calculated through constraint control, is retrieved. Based on the standardized process logic of new energy operation, and strictly following the pre-set sequence of procedures, operation execution timing logic, and node association relationships in the operation semantic path, all semantic unit nodes are ordered and linearly arranged. According to the sorting results, the semantic unit nodes are used as indices, and the corresponding post-propagation risk values ​​are numerical fields. The data is structured using an ordered array encapsulation method. The corresponding risk values ​​are entered sequentially according to the path node order. Redundant and invalid node data are removed, and it is ensured that the sequence nodes correspond one-to-one with the standard operation semantic path without errors or misalignments. Finally, a complete, continuous, time-consistent, and dimensionally regular ordered numerical sequence of post-propagation risk values ​​is constructed.

[0043] Based on the constructed ordered sequence of post-propagation risk values, the risk values ​​of two adjacent semantic unit nodes in the sequence are extracted sequentially. The risk change between nodes is calculated by subtracting the post-propagation risk value of the preceding node from the post-propagation risk value of the subsequent node. All adjacent node pairs along the entire path are continuously traversed to generate a complete original sequence of risk increments between nodes in batches. The state constraint threshold, operating boundary limit, and working condition constraint parameters bound to each semantic unit node are retrieved. A composite normalization algorithm combining extreme value truncation and range standardization is used. First, the original risk increment is truncated and corrected according to the upper and lower limits of the state constraints to remove abnormal and noise data. Then, the corrected increment value is linearly scaled using the range normalization formula to unify the units and numerical distribution range, eliminating data deviations caused by differences in the operating conditions and constraints of different nodes. Finally, a standardized risk increment that can accurately measure the dynamic evolution amplitude and intensity of risk under the premise of fluctuations in unit operating conditions and changes in operating state is obtained, ensuring the accuracy and comparability of subsequent risk accumulation calculations.

[0044] Based on the standardized risk increments arranged along the semantic path, a fixed-size continuous semantic unit sliding calculation window is set. The window is moved sequentially backward grid by grid according to the path flow. Within the coverage of each sliding window, all standardized risk increments within the window are accumulated and summed point by point to complete the full-domain segmented sliding accumulation calculation. The cumulative calculation results and window position information of each round of window are collected simultaneously. With the semantic unit node sequence as the horizontal axis and the risk accumulation value as the vertical axis, continuous interpolation fitting is performed to accurately construct a risk accumulation change curve that fits the path evolution law. Throughout the window sliding traversal and accumulation calculation process, the positive and negative attributes and the stage growth rate of the cumulative increment within the interval are monitored in real time. The continuous node intervals are checked segment by segment to see if they simultaneously meet the dual judgment conditions that the cumulative increment remains positive throughout and the growth rate within the unit node interval is monotonically increasing. The continuous node intervals that fully meet the constraints are uniformly screened, locked and marked separately as risk accumulation segments, realizing the accurate positioning and division of intervals where risks continuously accumulate and gradually increase.

[0045] The system accurately locates the terminal semantic unit node corresponding to each risk accumulation segment, extracts the actual risk value generated after propagation of the node, selects multiple sets of preceding node risk sequences arranged consecutively before the terminal node in the segment as basic samples, uses the least squares method to perform linear fitting on the preceding node risk data, establishes a time-series trend equation and completes trend extrapolation calculation to obtain the theoretical predicted risk value of the terminal node; firstly, the absolute deviation is calculated using the difference formula: absolute deviation = actual risk value after propagation - predicted risk value, and then the relative deviation is calculated by combining the predicted risk value: relative deviation = absolute deviation ÷ predicted risk value × 100%. The absolute deviation and relative deviation are merged and assigned a value, and the two quantitative results are combined as the final deviation degree, which intuitively and accurately quantifies the magnitude of the deviation of the actual risk data of the terminal node from the normal evolution trend.

[0046] The calculated node deviation is compared one by one with the fixed deviation thresholds pre-calibrated and stored in the system. Condition verification is carried out using threshold boundary judgment rules. When the value of the quantified deviation of a node is greater than the upper limit of the preset deviation threshold, the end node of the risk accumulation segment is immediately determined as the risk release trigger node. Using the locked risk release trigger node as the starting position of the reverse retrieval, the system traverses and backtracks node by node in the reverse order of the running semantic path to completely replicate and restore the entire risk accumulation segment to which the current node belongs, and locks the distribution range, number of nodes and interval boundaries of all continuous semantic unit nodes in the segment. Combining the increase characteristics of the standardized risk increment within the segment, the abrupt change characteristics of the risk accumulation curve and the influence range of state constraints, the interval locked by backtracking is calibrated and the range is corrected. Finally, the system accurately defines the local sudden increase segment of the risk value abnormal rise and concentrated outbreak within the process, and completes the accurate division and location of the abnormal risk interval.

[0047] In one possible implementation, step S300 further includes: Step S310: Perform structured parsing on the assessment request, extract job type tags, operation scenario tags and capability dimension constraint information, and construct a semantic constraint vector based on the capability dimension constraint information. The semantic constraint vector is used to characterize the responsibility boundary of the target job in the operation of new energy centralized control.

[0048] Step S320: Based on the semantic constraint vector, perform consistency scoring on each running semantic path in the running semantic path, and calculate the path fit index by combining the risk weight distribution of semantic units in the path.

[0049] Step S330: Sort and filter the operational semantic paths according to the path adaptability index, job type label, and operation scenario label. Among the candidate paths that meet the preset adaptability threshold, determine the target operational semantic path by combining the risk propagation structure integrity and key semantic unit coverage of the path.

[0050] Specifically, the system receives raw assessment request text data uploaded from external sources. It employs a domain-specific text structured parsing algorithm to segment, extract entities, and match keywords in the request content. This process precisely breaks down and independently extracts job type tags, operational scenario tags, and capability dimension constraints that define the assessment scope. For the extracted capability dimension constraints, the system first performs text feature vectorization. Through one-hot encoding combined with domain-specific word embedding mapping, different capability constraint items are transformed into standardized numerical feature components. Then, based on the weights and responsibilities of the new energy centralized control operation position, the components are weighted and the dimensions are concatenated to generate a semantic constraint vector with fixed dimensions and regular values. This semantic constraint vector, through the quantified numerical values ​​of different feature dimensions, accurately delineates and represents the target position's responsibilities, work authority boundaries, and core capability assessment boundaries in business scenarios such as new energy centralized control operation, status monitoring, anomaly handling, and risk assessment. This provides a standardized vector constraint basis for the accurate matching and scope limitation of subsequent assessment content.

[0051] Using the output semantic constraint vector as the benchmark for job responsibility determination, each operational semantic path is first represented globally by semantic vectorization. Then, the similarity between the semantic vector and the semantic constraint vector of each path is calculated using a cosine similarity algorithm. The corresponding semantic consistency score is obtained according to the similarity interval rules, thus measuring the degree to which the business content of the operational semantic path aligns with the target job's capability constraints and responsibilities. Next, all semantic unit nodes under the current operational semantic path are traversed, and the preset risk weight values ​​of each node are collected. The average risk weight within the path, the proportion of high-risk weight nodes, and the weight dispersion coefficient are calculated to fully characterize the risk weight distribution characteristics of a single path. The path fit index is quantitatively calculated using a two-dimensional weighted combination method. First, the semantic consistency score is normalized and scaled to a uniform numerical range. Then, the risk weight distribution characteristics are standardized without dimensions. Two fixed coefficients, semantic matching weight and risk structure weight, are preset. The fusion calculation is completed by the weighted calculation formula: Path Fit = Normalized Consistency Score × Semantic Matching Weight + Standardized Risk Weight Distribution Characteristics × Risk Structure Weight. At the same time, a scenario correction coefficient is introduced to calibrate the deviation. Finally, a quantifiable and comparable path fit index is accurately obtained, which objectively reflects the comprehensive fit level of each operating semantic path in terms of semantic compliance, risk structure distribution, and evaluation constraints.

[0052] Using the path adaptability index corresponding to each operational semantic path as the main ranking weight, and matching job type tags and operational scenario tags for precise association filtering, irrelevant paths with mismatched tags are first eliminated, and then the paths are sorted in descending order of path adaptability index from high to low. Then, all the sorted operational semantic paths are compared with the system's preset adaptability threshold, and paths with adaptability index greater than the preset adaptability threshold are selected as candidate paths. Next, the topological integrity of each candidate path is verified one by one. The structural integrity is determined by detecting the node connectivity of the risk propagation structure and the closed loop of the semantic links before and after. At the same time, the number of core key semantic units contained in the candidate paths and the overall coverage ratio are counted to quantify the key semantic unit coverage. Finally, within the candidate path range, based on the constraints of no breakpoints in the risk propagation structure and closed loop of the links, and combined with the key semantic unit coverage from high to low, the best candidates are selected and locked within the compliant candidate path range. The items with complete structure, comprehensive key content coverage and optimal adaptability are finally determined, and the target operational semantic path required for this evaluation is determined.

[0053] In one possible implementation, step S300 further includes: Step S340: Based on the dependency relationship and risk propagation structure of each semantic unit in the target running semantic path, perform key semantic unit perturbation selection. The perturbation selection includes determining the perturbation target node based on the risk weight and dependency centrality of the semantic unit.

[0054] Step S350: Perform operation information deletion disturbance, state constraint offset disturbance, and risk propagation relationship reconstruction disturbance on the disturbed target node to complete the disturbance processing.

[0055] Specifically, the algorithm analyzes the semantic path topology network of the target operation, sorts out the logical dependencies between all semantic units within the path, and extracts the node connections and link flow relationships of the overall risk propagation topology. Based on the pre-assigned risk weights of each semantic unit as the basic feature, a complex network topology algorithm is introduced to calculate the dependency centrality of each semantic unit. The degree centrality and betweenness centrality are jointly weighted to quantify the hub degree and core status of each node in the overall risk propagation link. The normalized risk weight score and the normalized dependency centrality score of the semantic unit are weighted and fused to generate a node comprehensive importance score. The nodes are sorted from high to low according to the comprehensive importance score, and the core semantic units with high scores and the greatest impact on the overall process and risk propagation are selected as the perturbation target nodes for subsequent interference, so as to achieve accurate perturbation selection of key semantic units based on structural features and risk attributes.

[0056] For the identified target semantic unit nodes, refined perturbation processing is implemented sequentially across three dimensions: When performing operational information deletion perturbation, the core business operation information corresponding to the node, such as centralized control operation steps, work instructions, and process execution details, is directly cleared, retaining the node framework but stripping away the actual operation content; when performing state constraint offset perturbation, the preset operating thresholds, upper and lower limits of operating conditions, and state judgment conditions of the node are quantitatively offset and adjusted, increasing or decreasing one or more constraint boundary values ​​according to a preset offset ratio to create deviations in normal state judgment; when performing risk propagation relationship reconstruction perturbation, the original risk propagation links between the target node and upstream and downstream semantic units are disconnected, the topological associations, transmission weights, and flow paths between nodes are rewritten, and the original risk propagation structure logic is rewritten; by simultaneously completing targeted artificial perturbations at the operational information, state constraint, and propagation topology levels, comprehensive structured perturbation processing of the target node is achieved.

[0057] In one possible implementation, step S100 further includes: Step S110: The semantic unit set is hierarchically classified according to the type of operation object and the running function attribute to form a hierarchical semantic unit structure.

[0058] Specifically, the entire set of operational semantic units is first decomposed into attributes, and the two core identifier features corresponding to each semantic unit are extracted: the type of operation object and the operation function attribute. The type of operation object is used as the first-level hierarchical basis, and the units are divided into major categories according to the categories of new energy centralized control operation equipment and control objects. Then, the operation function attribute is used as the second-level classification basis. Under the same operation object level, the units are further classified according to business functions such as monitoring and inspection, operation control, abnormal alarm, fault handling, and risk assessment. A hierarchical tree-like organizational structure is adopted to collect and group the semantic units level by level, set hierarchical affiliation rules and classification standards at the same level, eliminate attribute overlap and redundancy and standardize hierarchical relationships. A multi-level subordinate architecture of top-level major categories, middle-level subcategories and bottom-level units is built from top to bottom, and finally a hierarchical semantic unit structure with clear hierarchy, regular attributes and division according to both object and function dimensions is formed.

[0059] In one possible implementation, step S400 further includes: Step S410: The large model is constructed by parsing, slicing, and vectorizing new energy technology documents to establish an evaluation knowledge base.

[0060] Specifically, the large-scale model first collects various technical documents and procedures related to new energy centralized control operation, equipment maintenance, and fault handling in batches. It then performs intelligent text segmentation and paragraph slicing based on business chapters, process flows, and knowledge point boundaries, eliminating invalid blanks, redundant content, and formatting noise. Next, it performs domain-specific word segmentation, keyword entity extraction, and semantic feature extraction on the standardized text fragments. High-precision vectorization encoding is then achieved using an industry-specific word embedding model, transforming each knowledge slice into a fixed-dimensional semantic vector. Subsequently, it archives and indexes data according to job categories, operating scenarios, and capability dimensions, linking and storing the original text, semantic vectors, and category tags to build a structured and searchable new energy assessment-specific knowledge base. The large-scale model then uses this assessment knowledge base as its underlying knowledge foundation, combining domain-specific instruction fine-tuning and semantic retrieval enhancement mechanisms to complete specialized adaptation construction, possessing exclusive capabilities such as accurate recognition, semantic matching, test question generation, and capability assessment deduction.

[0061] In one possible implementation, step S400 further includes: Step S420: Obtain the target user's historical test records.

[0062] Step S430: If the assessment request does not meet the preset requirements, then establish the target user's weak knowledge points based on historical test records and compile a test paper, and output the assessment questions.

[0063] Specifically, based on the unique identifier of the target user, the system accurately retrieves all the user's past test records from the backend database of the assessment system. This includes complete data such as information on each assessment session, the correctness of answers to each knowledge point, detailed scores for each type of question, assessment results for each ability dimension, answering time, and archived information for incorrect questions. The original records are then structured and sorted chronologically to form a standardized user historical test dataset.

[0064] The system performs rule validation on the current assessment request to determine whether it meets preset requirements such as assessment specifications, job suitability, and scenario limitations. If the validation does not meet the preset requirements, it retrieves the target user's historical test records and accurately summarizes the target user's weak knowledge points by statistically analyzing the frequency of point deductions in each knowledge module, the concentrated distribution of incorrect questions, and the weak points in ability dimensions. Then, using the weak knowledge points as the core basis for question selection, it matches questions with corresponding question types, difficulty levels, and assessment dimensions from the assessment knowledge base. The system automatically completes intelligent test paper generation according to a reasonable test paper structure, knowledge point coverage, and difficulty ratio, and finally generates and outputs customized assessment questions that fit the user's ability weaknesses.

[0065] Example 2 is based on the same inventive concept as the new energy centralized control post evaluation and management method based on a large model in the previous examples, such as... Figure 2 As shown, this application provides a new energy centralized control job assessment and management system based on a large model. The system and method embodiments in this application are based on the same inventive concept. The system includes: The document parsing module 10 is used to parse the technical documents for new energy operation, extract the operation objects, operation actions and state constraint information, and convert the extraction results into quantifiable operation state feature vectors to construct a set of operation semantic units that include risk labeling and dependency relationships.

[0066] The risk propagation structure generation module 20 is used to construct the operational semantic path based on the temporal and dependency relationships of the set of operational semantic units, and to assign risk weights and propagation coefficients to the operational semantic path according to risk labeling to form a risk propagation structure.

[0067] The structured evaluation semantic establishment module 30 is used to construct a target running semantic path from the running semantic path after reading the evaluation request, and to establish structured evaluation semantics after performing perturbation processing on the target running semantic path.

[0068] The structured assessment semantic output module 40 is used to input structured assessment semantics into the large model and output assessment questions.

[0069] The weighted scoring module 50 is used to obtain the answer path corresponding to the assessment question, and then perform weighted scoring using the semantic path and risk propagation structure to output the ability assessment results and improvement suggestions.

[0070] Furthermore, the system is also used to implement the following functions: The response path is aligned with the execution semantic path. Based on the sequential consistency and dependency matching degree of semantic units, the effective matching sub-paths in the response path are located, and unmatched semantic units are marked as abnormal nodes. The consistency comparison of the response path is performed using the execution state feature vectors corresponding to each semantic unit to establish a state deviation. After mapping the abnormal nodes and state deviation to the risk propagation structure, risk accumulation propagation analysis is performed along the propagation coefficient of the semantic path to establish a risk deviation index. The coverage of effective matching sub-paths and the risk deviation index are obtained, and the operation step integrity score, state judgment accuracy score, and risk response consistency score are calculated to complete the weighted scoring.

[0071] Furthermore, the system is also used to implement the following functions: The abnormal nodes and state deviations are projected onto the corresponding semantic unit nodes in the risk propagation structure. The risk weights of each semantic unit node are reconstructed based on the type of the abnormal node and the state deviation, generating a dynamic risk propagation substructure that includes node risk increments and edge propagation sensitivity. The edge propagation sensitivity is constructed based on the strength of the dependency relationship between adjacent semantic units and the degree of influence of state deviations on subsequent semantic units. In the dynamic risk propagation substructure, the reconstructed node risk increments serve as the propagation source. Layered propagation calculations are performed along the semantic path using the propagation coefficients and the edge propagation sensitivity. During propagation, each semantic unit node receives propagation risks from upstream nodes, and the propagation risks are absorbed, amplified, or truncated based on the state constraint information of the corresponding nodes to obtain the post-propagation risk value of each semantic unit node. Based on the post-propagation risk value of each semantic unit node, the risk accumulation gradient change characteristics and local surge segments are extracted along the running semantic path. A risk deviation index reflecting the diffusion trend of answer deviations in the running process and key impact segments is constructed based on the extraction results.

[0072] Furthermore, the system is also used to implement the following functions: Arrange the semantic unit nodes in the operational semantic path according to the execution order to construct the corresponding post-propagation risk value sequence; calculate the risk increment sequence between nodes based on the post-propagation risk values ​​of adjacent semantic unit nodes, and perform constraint normalization processing on the risk increment sequence using state constraint information to obtain a standardized risk increment reflecting the degree of risk evolution under changes in unit operating conditions; perform sliding accumulation calculation on the standardized risk increment along the operational semantic path to construct a risk accumulation change curve within a continuous semantic unit window, and identify continuous node intervals that satisfy the condition that the cumulative increment is continuously positive and the growth rate is monotonically increasing during the sliding process, and mark them as risk accumulation segments; at the end node of the risk accumulation segment, calculate the degree of deviation between the actual post-propagation risk value and the predicted risk value obtained by extrapolating the risk of the preceding node; if the degree of deviation exceeds a preset deviation threshold, determine the corresponding node as a risk release trigger node, and backtrack the risk accumulation segment from the corresponding node to determine the local surge segment.

[0073] Furthermore, the system is also used to implement the following functions: The assessment request is structured and parsed to extract job type tags, operational scenario tags, and capability dimension constraint information. A semantic constraint vector is constructed based on the capability dimension constraint information, which represents the responsibility boundary of the target job in the centralized control operation of new energy. Based on the semantic constraint vector, the consistency score of each operational semantic path in the operational semantic path is calculated, and the path fit index is calculated in combination with the risk weight distribution of semantic units in the path. The operational semantic paths are sorted and filtered according to the path fit index, job type tags, and operational scenario tags. Among the candidate paths that meet the preset fit threshold, the target operational semantic path is determined by combining the risk propagation structure integrity and key semantic unit coverage of the path.

[0074] Furthermore, the system is also used to implement the following functions: Based on the dependencies and risk propagation structure of each semantic unit in the target running semantic path, key semantic unit perturbation selection is performed. The perturbation selection includes determining the perturbation target node based on the risk weight and dependency centrality of the semantic unit; and performing operation information deletion perturbation, state constraint offset perturbation, and risk propagation relationship reconstruction perturbation on the perturbation target node to complete the perturbation processing.

[0075] Furthermore, the system is also used to implement the following functions: The semantic unit set is hierarchically classified according to the type of operation object and the attributes of the operation function to form a hierarchical semantic unit structure.

[0076] Furthermore, the system is also used to implement the following functions: The large model is constructed by parsing, slicing, and vectorizing new energy technology documents to establish an evaluation knowledge base.

[0077] Furthermore, the system is also used to implement the following functions: Obtain the target user's historical test records; if the assessment request does not meet the preset requirements, then establish the target user's weak knowledge points based on the historical test records, compile a test paper, and output the assessment questions.

[0078] It should be noted that the order of the embodiments described above is for descriptive purposes only and does not represent the superiority or inferiority of the embodiments. Specific embodiments of this specification have been described above. Furthermore, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0079] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0080] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A method for evaluating and managing new energy centralized control positions based on a large model, characterized in that, The method includes: The technical documents for new energy operation are parsed to extract information on the operation objects, operation actions and state constraints. The extracted results are then transformed into quantifiable operation state feature vectors, and a set of operation semantic units containing risk labeling and dependency relationships is constructed. Based on the temporal and dependency relationships of the set of runtime semantic units, a runtime semantic path is constructed. Risk weights and propagation coefficients are assigned to the runtime semantic path according to risk labeling to form a risk propagation structure. After reading the evaluation request, a target runtime semantic path is constructed from the runtime semantic path matching, and after performing perturbation processing on the target runtime semantic path, a structured evaluation semantic is established. The structured assessment semantics are input into the large model, and the assessment questions are output. After obtaining the answer path corresponding to the assessment question, weighted scoring is performed using the semantic path of execution and the risk propagation structure, and the ability assessment results and improvement suggestions are output. After obtaining the answer path corresponding to the assessment questions, weighted scoring is performed using the semantic path of execution and the risk propagation structure, including: Align the answer path with the running semantic path, locate the valid matching sub-path in the answer path based on the sequential consistency of semantic units and the degree of dependency matching, and mark the unmatched semantic units as abnormal nodes. Consistency comparison of the response path is performed using the running state feature vectors corresponding to each semantic unit to establish the state deviation quantity; After mapping abnormal nodes and state deviations to the risk propagation structure, risk accumulation propagation analysis is performed along the propagation coefficients of the semantic path to establish a risk deviation index. Obtain the coverage and risk deviation indicators of the effective matching sub-paths, and calculate the operation step integrity score, status judgment accuracy score, and risk response consistency score to complete the weighted scoring; After mapping abnormal nodes and state deviations to the risk propagation structure, risk accumulation and propagation analysis is performed along the propagation coefficients of the semantic path, including: The abnormal nodes and state deviations are projected onto the corresponding semantic unit nodes in the risk propagation structure. The risk weights of each semantic unit node are reconstructed based on the type of the abnormal node and the state deviation, generating a dynamic risk propagation substructure that includes node risk increment and edge propagation sensitivity. The edge propagation sensitivity is constructed based on the strength of the dependency relationship between adjacent semantic units and the degree of influence of state deviation on subsequent semantic units. In the dynamic risk propagation substructure, the reconstructed node risk increment is used as the propagation source. The propagation coefficient along the semantic path and the edge propagation sensitivity are used to perform hierarchical propagation calculation. During the propagation process, the propagation risk received from the upstream node is accumulated for each semantic unit node. The propagation risk is absorbed, amplified or truncated according to the state constraint information of the corresponding node to obtain the post-propagation risk value of each semantic unit node. Based on the risk value after propagation of each semantic unit node, the risk accumulation gradient change characteristics and local sudden increase segments are extracted along the running semantic path. Based on the extraction results, a risk deviation index reflecting the diffusion trend of answer deviation in the running process and key impact segments is constructed.

2. The new energy centralized control post evaluation and management method based on a large model as described in claim 1, characterized in that, Based on the post-propagation risk value of each semantic unit node, the cumulative risk gradient change features and local sudden increase segments are extracted along the running semantic path, including: Arrange the semantic unit nodes in the execution semantic path according to the execution order to construct the corresponding post-propagation risk value sequence; The risk increment sequence between nodes is calculated based on the risk value after propagation of adjacent semantic unit nodes. The risk increment sequence is then subjected to constraint normalization processing using state constraint information to obtain a standardized risk increment that reflects the degree of risk evolution under changes in unit operating conditions. Perform sliding accumulation calculation on the standardized risk increment along the semantic path of operation, construct the risk accumulation change curve within the continuous semantic unit window, and identify continuous node intervals that satisfy the condition that the cumulative increment is continuously positive and the growth rate is monotonically increasing during the sliding process, and mark them as risk accumulation segments. At the end node of the risk accumulation section, the degree of deviation between the actual risk value after propagation and the predicted risk value obtained by extrapolating the risk from the preceding node is calculated. If the deviation exceeds the preset deviation threshold, the corresponding node will be identified as a risk release trigger node, and the risk accumulation section will be traced back from the corresponding node to determine the local surge section.

3. The new energy centralized control post evaluation and management method based on a large model as described in claim 1, characterized in that, After reading the evaluation request, the target runtime semantic path is constructed from the runtime semantic path matching, including: The assessment request is structured and parsed to extract job type tags, operation scenario tags, and capability dimension constraint information. A semantic constraint vector is constructed based on the capability dimension constraint information. The semantic constraint vector is used to characterize the responsibility boundary of the target job in the operation of new energy centralized control. Based on the semantic constraint vector, a consistency score is given for each semantic path in the operation semantic path, and a path fit index is calculated by combining the risk weight distribution of semantic units in the path. The operational semantic paths are sorted and filtered according to the path adaptability index, job type tag, and operation scenario tag. Among the candidate paths that meet the preset adaptability threshold, the target operational semantic path is determined by combining the risk propagation structure integrity and key semantic unit coverage of the path.

4. The new energy centralized control post evaluation and management method based on a large model as described in claim 1, characterized in that, Perturbation processing of the target execution semantic path includes: Based on the dependency relationships and risk propagation structure of each semantic unit in the target running semantic path, a key semantic unit perturbation selection is performed. The perturbation selection includes determining the perturbation target node based on the risk weight and dependency centrality of the semantic unit. To complete the disturbance handling, the perturbation target node is subjected to operation information deletion perturbation, state constraint offset perturbation, and risk propagation relationship reconstruction perturbation.

5. The new energy centralized control post evaluation and management method based on a large model as described in claim 1, characterized in that, The semantic unit set is hierarchically classified according to the type of operation object and the attributes of the operation function to form a hierarchical semantic unit structure.

6. The new energy centralized control post evaluation and management method based on a large model as described in claim 1, characterized in that, The large model is constructed by parsing, slicing, and vectorizing new energy technology documents to establish an evaluation knowledge base.

7. The new energy centralized control post evaluation and management method based on a large model as described in claim 1, characterized in that, The structured assessment semantics are input into the large model, and the output includes assessment items, which also include: Obtain the target user's historical test records; If the assessment request does not meet the preset requirements, the system will identify the target user's weak knowledge points based on historical test records, compile a test paper, and output the assessment questions.

8. A new energy centralized control post evaluation and management system based on a large model, characterized in that: The system is used to implement the new energy centralized control post assessment and management method based on a large model as described in any one of claims 1-7, and the system includes: The document parsing module is used to parse new energy operation technical documents, extract operation objects, operation actions and state constraint information, and transform the extraction results into quantifiable operation state feature vectors to construct a set of operation semantic units that include risk labeling and dependency relationships. The risk propagation structure generation module is used to construct the operational semantic path based on the temporal and dependency relationships of the set of operational semantic units, and to assign risk weights and propagation coefficients to the operational semantic path according to risk labeling to form a risk propagation structure. The structured evaluation semantic establishment module is used to construct a target running semantic path from the running semantic path after reading the evaluation request, and to establish the structured evaluation semantic after performing perturbation processing on the target running semantic path. The structured assessment semantic output module is used to input structured assessment semantics into the large model and output assessment questions. The weighted scoring module is used to obtain the answer path corresponding to the assessment questions, and then perform weighted scoring using the semantic path and risk propagation structure to output the ability assessment results and improvement suggestions.

Images