Coal quality data processing method, device, equipment, medium and program

Abstract

This application relates to the field of data processing technology, and in particular to a method, apparatus, equipment, medium, and program for processing coal quality data. The method includes: acquiring coal quality data from different data sources and converting the coal quality data into coal quality feature vectors; inputting each coal quality feature vector into an anomaly detection model, which outputs a first verification result for the corresponding coal quality data; inputting the coal quality feature vectors into a coal calorific value prediction model, which outputs a predicted calorific value for the coal quality data; generating a second verification result based on the predicted calorific value and a preset threshold; and generating a verification report based on the first and second verification results of the coal quality data. This solves the problems in related technologies where relying on manual experience and single rules to verify coal quality data leads to missed anomalies, large prediction deviations, and difficulty in meeting real-time requirements.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method, apparatus, equipment, medium and program for coal quality data processing. Background Technology

[0002] As key targets for carbon emission control, the accuracy of carbon emission accounting for power generation enterprises highly depends on the reliability of coal quality data. Coal quality data is the core input parameter for calculating the carbon dioxide emission factor and emission volume of coal combustion. However, in reality, coal quality data often comes from multiple heterogeneous sources, including coal mine factory test reports, power plant incoming coal testing, real-time monitoring of coal entering the furnace, and data issued by third-party testing agencies. These data differ significantly in terms of collection standards, testing methods, baseline conditions, field naming conventions, and units of measurement, resulting in severe data silos and making it difficult to form a unified, authoritative, and traceable high-quality coal quality database.

[0003] In related technologies, the verification of coal quality data mainly relies on manual experience or outlier removal methods based on simple statistical rules. For example, box plots are often used to set fixed thresholds to identify outliers, or logical consistency checks are performed based on the physicochemical constraints between coal components. Although such methods are effective in handling obviously erroneous data, they are inadequate when dealing with complex coal quality data that is high-dimensional, nonlinear, and multimodal. These methods suffer from weak multi-source data fusion capabilities, limited and robust anomaly detection methods, insufficient accuracy and weak generalization ability in calorific value prediction, and inefficient manual review processes. Consequently, they fail to meet the stringent requirements of the carbon trading market for real-time and accurate data. Summary of the Invention

[0004] This application provides a coal quality data processing method, apparatus, equipment, medium, and program to solve the problems in related technologies that rely on manual experience and single rules to verify coal quality data, resulting in missed anomalies, large prediction deviations, and difficulty in meeting real-time requirements.

[0005] The first aspect of this application provides a coal quality data processing method, comprising the following steps: acquiring coal quality data from different data sources and converting the coal quality data into coal quality feature vectors; inputting each coal quality feature vector into an anomaly detection model, wherein the anomaly detection model outputs a first verification result corresponding to the coal quality data, wherein the anomaly detection model collaboratively calculates the confidence level of the coal quality data through at least one sub-model, and generates a corresponding first verification result based on the confidence level; inputting the coal quality feature vector into a coal calorific value prediction model, wherein the coal calorific value prediction model outputs a predicted calorific value of the coal quality data, and generates a second verification result based on the predicted calorific value and a preset threshold; and generating a verification report based on the first verification result and the second verification result of the coal quality data.

[0006] Optionally, before converting the coal quality data into a coal quality feature vector, the method includes: extracting the sample information field and the coal quality information field from each record in the coal quality data after field mapping and benchmark conversion; generating a key-value sequence based on the sample information field and the coal quality information field; calculating the corresponding hash value based on the string concatenated from the key-value sequence; removing duplicate coal quality data records based on the hash value; and converting the deduplicated coal quality data into a coal quality feature vector.

[0007] Optionally, the anomaly detection model includes: a density-based hierarchical clustering sub-model, an isolated forest sub-model, a single-class support vector machine, and a Gaussian mixture sub-model. The density-based hierarchical clustering sub-model calculates the core distance and inter-sample reach distance for each sample point based on the coal quality feature vector, constructs a density hierarchy based on the core distance and inter-sample reach distance, outputs noise labels and outlier scores, and generates a first anomaly score based on the noise labels and outlier scores. The isolated forest sub-model constructs multiple random binary trees and calculates the average path length of the sample in the forest, generating a second anomaly score based on the average path length. The single-class support vector machine maps the coal quality feature vector to a high-dimensional space using a kernel function, uses a decision function to determine the relationship between the sample and the boundary, outputs a boundary distance score, and generates a third anomaly score based on the boundary distance score. The Gaussian mixture sub-model estimates the parameters of the Gaussian mixture model using an expectation-maximization algorithm to fit the global probability distribution of the sample, calculates the negative log-likelihood of the sample, and generates a fourth anomaly score based on the negative log-likelihood.

[0008] Optionally, the processing method of the anomaly detection model includes: normalizing the first anomaly score, the second anomaly score, the third anomaly score, and the fourth anomaly score to generate corresponding confidence scores; obtaining the number of hard votes for each sub-model; calculating a comprehensive confidence score based on the confidence scores and their respective weights; and determining the first verification result of the coal quality data based on the comprehensive confidence score and the number of hard votes.

[0009] Optionally, the processing method of the coal calorific value prediction model includes: calculating the Euclidean distance between each coal quality feature vector and each cluster centroid, determining the nearest cluster subgroup based on the Euclidean distance; calling the regression model corresponding to the nearest cluster subgroup; inputting the coal quality feature vector of the sample to be predicted into the regression model, and the regression model outputting the predicted calorific value of the coal quality data.

[0010] Optionally, the training method for the coal calorific value prediction model includes: dividing coal quality data with similar compositional structures into multiple clusters; within each cluster, fitting a regression model between the received basis lower calorific value and coal quality composition indicators, wherein candidate models are generated by repeatedly randomly sampling multiple samples from the coal quality data sample set, calculating the residual value of each sample in the coal quality data sample set, determining the inliers based on the residual value and a preset threshold to generate an inlier set, selecting the candidate model with the largest number of inliers, refitting the candidate model based on the inliers to obtain the final regression parameters, and updating the regression model based on the final regression parameters.

[0011] A second aspect of this application provides a coal quality data processing apparatus, comprising: an acquisition module for acquiring coal quality data from different data sources and converting the coal quality data into coal quality feature vectors; a first processing module for inputting each coal quality feature vector into an anomaly detection model, wherein the anomaly detection model outputs a first verification result corresponding to the coal quality data, wherein the anomaly detection model collaboratively calculates the confidence level of the coal quality data through at least one sub-model and generates a corresponding first verification result based on the confidence level; a second processing module for inputting the coal quality feature vectors into a coal calorific value prediction model, wherein the coal calorific value prediction model outputs a predicted calorific value of the coal quality data and generates a second verification result based on the predicted calorific value and a preset threshold; and a generation module for generating a verification report based on the first verification result and the second verification result of the coal quality data.

[0012] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to perform the coal quality data processing method as described in the above embodiments.

[0013] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to perform the coal quality data processing method as described in the above embodiments.

[0014] A fifth aspect of this application provides a computer program product, including a computer program or instructions, which, when executed, implement the coal quality data processing method as described in the above embodiments.

[0015] Therefore, this application has at least the following beneficial effects: This application embodiment integrates multi-source coal quality data and transforms it into a standardized coal quality feature vector. First, an anomaly detection model, composed of multiple sub-models working collaboratively, assesses the confidence level of each data point, generating a first verification result reflecting the overall rationality of the data. Simultaneously, the same feature vector is input into a specially constructed coal calorific value prediction model to obtain a calorific value prediction based on component correlation. This prediction is then combined with a preset deviation threshold to determine the credibility of the measured calorific value, forming a second verification result. Finally, the two verification results are merged to generate a comprehensive verification report. This allows for multi-dimensional intelligent verification of the logical consistency, physical rationality, and numerical reliability of coal quality data without the need for real labels, significantly improving the quality and credibility of the basic data upon which carbon emission accounting relies.

[0016] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0017] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart of a coal quality data processing method according to an embodiment of this application; Figure 2 This is a flowchart of a coal quality data processing method according to an embodiment of this application; Figure 3 This is a block diagram of a coal quality data processing apparatus provided according to an embodiment of this application; Figure 4 This is a schematic diagram of the coal quality data processing architecture provided according to an embodiment of this application; Figure 5 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0018] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0019] The coal quality data processing method, apparatus, equipment, medium, and program of this application are described below with reference to the accompanying drawings.

[0020] Specifically, Figure 1 This is a flowchart illustrating a coal quality data processing method provided in an embodiment of this application.

[0021] like Figure 1As shown, the coal quality data processing method includes the following steps: In step S101, coal quality data from different data sources are obtained, and the coal quality data is converted into coal quality feature vectors.

[0022] It is understood that the embodiments of this application can acquire coal quality data from different data sources and convert the coal quality data into coal quality feature vectors, thereby eliminating the problem of data incomparability caused by differences in sources, different units or chaotic benchmarks, and providing a structured and high-quality input basis for subsequent anomaly detection and calorific value prediction.

[0023] It should be noted that data sources can include: coal received from power plants, coal fed into furnaces, coal mine exit reports, and third-party testing. Coal quality data can include key coal quality indicators such as received basis net calorific value, total moisture, dry basis total sulfur, and dry basis ash content. Because coal quality data from different data sources differ significantly in format, units, baseline conditions (e.g., received basis a, air-dried basis ad, dry ash-free basis daf), sampling time granularity, and field naming conventions, a multi-level data preprocessing mechanism is adopted: firstly, missing values, obvious input errors, or unit confusion items are preliminarily cleaned; secondly, data is preprocessed according to coal analysis standards and... The thermodynamic benchmark conversion formula unifies all component indicators to a common benchmark, ensuring that parameters such as ash, sulfur, moisture, and calorific value are physically comparable. Then, through feature engineering, representative coal quality dimensions are extracted, including but not limited to core indicators such as net calorific value, total sulfur, ash, moisture, and dry ash-free volatile matter. Derivative features (such as estimated hydrogen content, fixed carbon approximation, or elemental ratios) are introduced according to business needs. Finally, all features are normalized or standardized to eliminate the influence of dimensions, forming a coal quality feature vector with consistent structure, stable values, and clear semantics.

[0024] Specifically, raw data from sources such as coal mine testing, incoming coal quality reports, and coal quality tracking are extracted and converted into structured records (one record corresponds to the testing results of a batch or a single coal sample). A standardized naming set, unit benchmarks, and metadata for the coal quality fields required in this application are established using a data dictionary. This specifies the Chinese names / symbols of the fields, benchmark identifiers (ar / d / ad), units of measurement, data types, reasonable ranges, mandatory fields, and synonyms.

[0025] The data naming dictionary (standard data fields) in this application shall include at least the following: number, test date, coal type, product name, coal sample type, mine source location, supplier, whether it is blended, shipping platform, industry type, data reporting unit, batch coal quantity (t), and total moisture content. (%), moisture content on air-dried basis (%) Received basic sulfur (%) Received basic carbon (kcal / kg), lower heating value on received basis (kcal / kg), dry ash content (%), volatile matter on dry basis (%), dry basis total sulfur (%), dry-based carbon (%), dry-based hydrogen (%), Higher heating value on dry basis (kcal / kg), deformation temperature (°C), softening temperature (°C), hemispherical temperature (°C), flow temperature (°C), Air-dried ash content (%), Volatile matter on air-dried basis (%), air-dried total sulfur (%), air-dried carbon (%), air-dried hydrogen (%), Higher heating value on air-dried basis (kcal / kg).

[0026] Establish mapping rules to map the original field names, common names, units of measurement, and baseline status in any data source to the set of rules for the above-mentioned standardized fields. These rules include: a synonym / alias mapping table (e.g., air-dried basis water is often called internal water), rule matching and fuzzy matching (matching field names based on word segmentation and edit distance / similarity), unit conversion and baseline conversion functions, and conflict handling priority (selecting based on data source credibility / timestamp / completeness when there are multiple values ​​for the same field).

[0027] The inconsistency in field naming is resolved by establishing a standardized data naming dictionary and combining it with an alias table and fuzzy matching. The inconsistency in indicator benchmarks is resolved by explicitly maintaining benchmark labels (received basis ar, dry basis d, empty dry basis ad) at the field level and performing benchmark conversion. For calorific value indicators, the database uniformly uses kcal / kg as the benchmark unit. When the benchmark unit corresponding to coal quality data is MJ / kg, it is converted to kcal / kg using the unit conversion formula Q(kcal / kg)=Q(MJ / kg)×238.8459 (using 1kcal=4.1868kJ) before being stored in the coal quality database.

[0028] Based on the above, all data is transformed into a unified structure and benchmark: the structure refers to a standardized wide table / relational table structure with standardized fields as columns (each row corresponds to one coal sample / batch record, and includes traceability fields such as data source, warehousing time, original field name, and original unit); the benchmark refers to a unified reporting benchmark used for carbon emission accounting and cross-source comparison, preferably a unified benchmark with the received benchmark as the key indicator (e.g., , (etc.), and can simultaneously retain or calculate derived benchmark fields such as dry basis d, air-dried basis ad for traceability.

[0029] In this embodiment of the application, before converting the coal quality data into a coal quality feature vector, the process includes: extracting the sample information field and the coal quality information field from each record in the coal quality data after field mapping and benchmark conversion; generating a key-value sequence based on the sample information field and the coal quality information field; calculating the corresponding hash value based on the string concatenated from the key-value sequence; removing duplicate records of coal quality data based on the hash value; and converting the deduplicated coal quality data into a coal quality feature vector.

[0030] It is understood that the embodiments of this application can achieve semantic alignment and physical consistency of multi-source coal quality data through field mapping and benchmark conversion. Then, sample information fields (such as sampling time, batch number, source system) and coal quality information fields (such as standardized indicators such as calorific value, sulfur content, ash content, etc.) are separated from each record. The two are combined into a structured key-value sequence and concatenated into a unique string. A hash algorithm is used to generate a digital fingerprint of the record, thereby efficiently identifying and eliminating data with completely duplicate content. The non-redundant coal quality records after deduplication are then converted into feature vectors with fixed dimensions and standardized values. This avoids model bias caused by data duplication and ensures the uniqueness and representativeness of the input features, thereby improving the accuracy and computational efficiency of subsequent anomaly detection and calorific value prediction.

[0031] It should be noted that the core logic of the complete data governance process for coal quality data, from integration to high-quality data storage, can be summarized in four stages: deduplication, completion, cleaning, and modeling. (1) Efficient deduplication mechanism based on hash fingerprint To eliminate duplicate records caused by multi-source collection (such as a batch of coal being reported multiple times), the system combines basic sample information (such as sampling time, batch number, and source) and coal quality information (such as standardized fields like calorific value and ash content) in each record into a key-value pair sequence. This sequence undergoes rigorous standardization, including removing redundant spaces, standardizing units and benchmarks, explicitly setting missing values ​​to null, and arranging fields in a fixed order—ensuring that semantically identical data generates completely consistent string representations. A unique digital fingerprint is then generated using the MD5 hash algorithm. During the data entry stage, a hash table is used for fast deduplication with minimal time complexity; when the data volume is extremely large, a pre-filter can be used for probabilistic pre-screening, significantly reducing storage and computational overhead. For duplicate records with the same hash value, only the one with the smaller primary key or earlier entry time is retained, while the rest are discarded, thus effectively controlling redundancy while ensuring data integrity.

[0032] (2) Intelligent missing value completion based on physical laws When a key coal quality indicator is missing under a certain benchmark but other benchmark data are complete, the system does not simply discard or interpolate it, but instead uses logical deduction to complete it based on the benchmark conversion principle of coal analysis.

[0033]

[0034] in, The mass fraction indicators include ash content, sulfur content, and volatile matter. For the moisture content under the corresponding benchmark (received from the base) Air-dried base dry base , The component content is based on bi. Moisture content at baseline bi.

[0035] It should be noted that, assuming the absolute dry matter mass of a batch of coal is a fixed value (e.g., 100 grams of dry coal), then a general conversion formula can be derived: ; ; ; in, It is total water content. Moisture content on an air-dried basis. To obtain the content under the base (ar), The content of components on a dry basis (d) The content of components on an empty dry basis (ad).

[0036] It should be noted that the specific conversion relationship of the coal quality benchmark is shown in Table 1 below, where M represents moisture and A represents the variable that needs to be converted to the benchmark.

[0037] Table 1

[0038] In addition, the conversion of heat output is as follows: ; ; in, The high calorific value is based on an air-dry basis. For the heat generation of the air-dry base cartridge, This represents the total sulfur content on an air-dried basis. This is the thermal correction factor for nitric acid formation in the heat of combustion of the bombardier. To receive the base low heat value, The high calorific value is based on an air-dry basis. The hydrogen content is based on the air dry basis. To obtain the base moisture, Moisture content is on an air-dried basis.

[0039] (3) Anomaly cleaning that integrates statistical rules and domain knowledge After deduplication and completion, a dual-validation strategy is employed to identify and remove outliers: Statistically, the 97.5th percentile is set as a dynamic threshold to capture extreme outliers, while reasonable ranges are defined based on coal physicochemical principles (e.g., lignite's calorific value is typically <5000 kcal / kg, and anthracite's ash melting point is >1350℃). Data is only deemed invalid and removed when it simultaneously violates statistical laws and physical feasibility, balancing robustness and professional considerations.

[0040] Specifically, data governance is performed on the integrated set of data: The sample basic information field and coal quality information field from each record are selected to form a key-value sequence. The key-value sequences are standardized (removing spaces, unifying units and benchmarks, setting missing values ​​to null, and sorting fields), concatenated into a string s, and its MD5 hash fingerprint h=MD5(s) is calculated and stored as a new field hash_md5. During data entry, a hash table is used for rapid deduplication (a Bloom filter can be used for pre-filtering when the data volume is extremely large). For records with the same hash_md5, only the one with the smaller sequence number or primary key (or an earlier entry time) is retained; the rest are considered duplicates and removed, thus achieving efficient deduplication and eliminating redundant records.

[0041] To address the issue of missing values ​​in coal quality data under a specific benchmark, if the corresponding moisture content (M) and other benchmarks provide complete coal quality data, the missing values ​​are calculated and supplemented based on the coal quality benchmark conversion relationship, ensuring that key indicators are complete and usable under a unified benchmark. After deduplication and supplementation, outlier cleaning is performed using statistical methods and coal industry knowledge (verified using a combination of 97.5% quantile and physically reasonable range), eliminating obviously unreasonable data points. Finally, a relational database is designed based on the entity relationships of the coal quality data, defining table structures for coal samples, sampling time, origin, and various analytical indicators, and the cleaned data is loaded into the database. The database supports efficient querying and flexible access, and ensures data security through access control. This high-quality coal quality database provides a solid data foundation for subsequent model training and validation applications.

[0042] In step S102, each coal quality feature vector is input into the anomaly detection model, and the anomaly detection model outputs the first verification result of the corresponding coal quality data. The anomaly detection model calculates the confidence level of the coal quality data collaboratively through at least one sub-model, and generates the corresponding first verification result based on the confidence level. It is understood that the embodiments of this application can input the standardized coal quality feature vector into an anomaly detection model composed of multiple sub-models. Each sub-model evaluates the rationality of the data in the multi-dimensional feature space based on different unsupervised learning mechanisms (such as isolated forest, local outlier, autoencoder, etc.), and generates a unified confidence score through weighted fusion or voting strategies. The confidence score reflects the consistency between the sample and the normal coal quality distribution pattern. When it is lower than a preset threshold, it is judged as an anomaly. Finally, the structured first verification result is output according to the confidence score, so as to achieve high sensitivity identification of suspicious coal quality data such as false reports, misrecording, or physical contradictions, and effectively improve the robustness and generalization ability of data quality verification.

[0043] In this embodiment, the anomaly detection model includes: a density-based hierarchical clustering sub-model, an isolated forest sub-model, a single-class support vector machine, and a Gaussian mixture sub-model. The density-based hierarchical clustering sub-model calculates the core distance and inter-sample reach distance for each sample point based on the coal quality feature vector, constructs a density hierarchy based on the core distance and inter-sample reach distance, outputs noise labels and outlier scores, and generates a first anomaly score based on the noise labels and outlier scores. The isolated forest sub-model constructs multiple random binary trees and calculates the average path length of the sample in the forest, generating a second anomaly score based on the average path length. The single-class support vector machine maps the coal quality feature vector to a high-dimensional space using a kernel function, uses a decision function to determine the relationship between the sample and the boundary, outputs a boundary distance score, and generates a third anomaly score based on the boundary distance score. The Gaussian mixture sub-model estimates the parameters of the Gaussian mixture model using an expectation-maximization algorithm to fit the global probability distribution of the sample, calculates the negative log-likelihood of the sample, and generates a fourth anomaly score based on the negative log-likelihood. It is understood that the anomaly detection model in this application integrates four complementary unsupervised sub-models to characterize the anomalous features of coal quality feature vectors from different perspectives: the HDBSCAN (Hierarchical Density-Based Spatial Clustering of Applications with Noise) sub-model, based on a density hierarchy, identifies local sparse regions and outputs noise labels and outlier scores through core distance and reach distance; the isolated forest sub-model utilizes a random segmentation mechanism to reflect the isolation degree of a sample based on the average path length of the sample in multiple binary trees, generating a second anomaly score; the single-class support vector machine maps the data to a high-dimensional feature space, measures the degree of deviation from the normal domain through the distance from the sample to the learning boundary, forming a third anomaly score; the Gaussian mixture model fits the global probability distribution through the expectation-maximization algorithm, quantifies the probability of occurrence of the sample using the negative log-likelihood, and generates a fourth anomaly score; finally, the four scores are fused into a comprehensive confidence score through a weighted or integrated strategy, which is used to determine whether the data is abnormal, thereby achieving multi-dimensional and highly robust identification of complex anomaly patterns such as false reporting, misrecording, or physical inconsistency in coal quality data.

[0044] Specifically, within an unsupervised learning framework, this application develops an anomaly detection model integrating multiple algorithms for automatically verifying the credibility of coal quality data. This module integrates four types of detection algorithms based on different principles: HDBSCAN, isolated forest sub-model, single-class support vector machine, and Gaussian mixture sub-model. Each sub-model assesses the data from different perspectives and provides a confidence score. To ensure reproducibility and implementability, the same data preprocessing is used uniformly during both the training and inference phases: key feature fields containing coal quality information are selected from the coal quality database (in this example, dry basis ash content, dry basis total sulfur, total moisture, and lower heating value on the received basis) to construct a d-dimensional feature vector x∈R. d (In this example, d=4); Perform Z-score standardization on the numerical features: z=(x u) / s, where u is the mean of the training set and s is the standard deviation of the training set, to obtain the normalized feature z; missing records that cannot be reliably completed are removed.

[0045] The working principle, calculation method, model structure, and function of each sub-model are as follows: (a) HDBSCAN: This model detects anomalies from the perspective of local density. For each sample point... Define core distance For its first Nearest neighbor distance; defines reach distance ,in The distance between samples is the Euclidean distance. A minimum spanning tree is constructed based on the reach distance to form a density hierarchy. The final cluster is selected based on cluster stability, and points that cannot be assigned to a stable cluster are marked as noise points. The model outputs cluster labels. With noise markers (Noise level is 1), and outlier scores can be provided. (For example, outlier degree based on hierarchical tree).

[0046] (b) Isolated Forest: This model identifies anomalies from the perspective of the difficulty of random isolation. It constructs T random binary trees, randomly selects a feature q for each tree, and randomly selects a split point p to recursively segment the sample space until a sample is isolated or the maximum depth is reached. The shorter the average path length E[h(x)] of a sample in the forest, the more likely it is to be an anomaly; its anomaly score can be calculated according to... Calculate, where c(n) is the average path length constant when the sample size is n.

[0047] (c) Single-class Support Vector Machine: This model learns the envelope of normal data from the perspective of the boundary / support domain. It maps samples to a high-dimensional space using a kernel function φ(·) to solve the optimization problem. , The constraints are: i、 ; derive the decision function Where w, ξ, and ρ are the weight vector, slack variable, and bias term, respectively, v is the adjustment parameter, and n is the total number of samples.

[0048] like If it is determined to be a point of contention, a vote will be generated. and boundary distance score ,in, The output value of the decision function for sample x (i.e., coal quality feature vector) in the high-dimensional feature space.

[0049] (d) Gaussian Mixture Submodel: This model identifies anomalies from a global probability density / likelihood perspective. Assume the data is generated by... Generated by a mixture of Gaussian components, with a probability density of... ,in For weight, For the mean, For covariance, the expectation-maximization algorithm is used to iteratively estimate the parameters (E-step to calculate posterior responsibility, M-step to update). , , (Maximize the log-likelihood). Calculate the negative log-likelihood for the sample. , The larger the value, the more the sample deviates from the main distribution, and the more questionable it is.

[0050] The four types of models described above characterize the suspicious aspects from different perspectives (local density, random isolation, boundary support region, and global probability density), thus forming a complementary approach. The confidence score is obtained as follows: each model first generates an initial anomaly score. (m∈{H,I,F,SVM,GMM}), and then the scores are uniformly mapped to the [0,1] interval to obtain the confidence level. .

[0051] One feasible mapping approach is quantile normalization: compute the empirical distribution F of the ratings on the training set (or its approximately normal subset). m ( ), make (Models with higher scores and more abnormal characteristics can be used directly) This unifies the scoring criteria across different models, making the scores closer to 1 more suspicious. F is the original anomaly score for the m-th sub-model. m ( Let be the cumulative distribution function of the anomaly score of the m-th sub-model. Let be the confidence level of the m-th sub-model for sample x.

[0052] Obtaining binary votes from each model With confidence level Subsequently, this invention designs a collaborative decision-making system that integrates hard voting and confidence-weighted evaluation: first, the number of hard votes is counted. Secondly, calculate the overall confidence level. ,in, These are the model weights (they can be equal by default, or determined according to model stability / consistency evaluation).

[0053] The final decision rule can be set as follows: If ( If it is, then it is deemed questionable; or when Less than half but (threshold) If the condition is not met, it is considered questionable; otherwise, it is considered normal.

[0054] To further reduce false positives and false negatives in unlabeled scenarios, this paper also provides a multi-model collaborative training mechanism based on HDBSCAN initialization: first, HDBSCAN is used to obtain the initial noise ratio. Based on this, the isolation forest contamination was established. One-Class SVM Parameters and GMM component count The candidate range is defined; then, fine-tuning is performed within this range through ensemble sampling and parallel search; the data set that is highly consistently judged as questionable by multiple models and has high overall confidence is defined as the high-confidence questionable set, which is removed from the training set and each model is retrained. Marginal samples that are judged as questionable by only a few models but considered normal by other models are retained, allowing different models to teach each other through consistency screening, thereby improving the robustness and interpretability of the ensemble judgment. During online runtime, after inputting new quality inspection data, the same field mapping, unit / benchmark conversion, and Z-score standardization are performed, and then fed into the four sub-models to obtain... , and It outputs the normal / questionable judgment, the overall confidence level, and the triggering basis of each model, ensuring that the results are traceable.

[0055] In this embodiment of the application, the processing method of the anomaly detection model includes: normalizing the first anomaly score, the second anomaly score, the third anomaly score, and the fourth anomaly score to generate corresponding confidence scores; obtaining the number of hard votes corresponding to each sub-model; calculating the comprehensive confidence score based on the confidence score and their respective weights; and determining the first verification result of the coal quality data based on the comprehensive confidence score and the number of hard votes.

[0056] It is understood that the embodiments of this application can normalize the anomaly scores output by the four sub-models respectively, and uniformly map them to a comparable confidence interval to reflect the strength of each model's judgment on the normality of the sample; at the same time, the hard voting results (i.e., the number of binary votes for normal or abnormal) generated by each sub-model based on a preset threshold are counted; then, the performance of each sub-model in historical verification is combined with corresponding weights, and the normalized confidence is weighted and fused to generate a comprehensive confidence; finally, the comprehensive confidence and the consistency of the hard voting are used to make a joint decision. If the majority of sub-models vote for anomaly and the comprehensive confidence is lower than the judgment threshold, the coal quality data is marked as anomaly; otherwise, it is considered reliable. This achieves a robust verification mechanism that takes into account both model consensus and confidence strength, and improves the accuracy and reliability of anomaly identification.

[0057] In step S103, the coal quality feature vector is input into the coal calorific value prediction model, the coal calorific value prediction model outputs the calorific value prediction value of the coal quality data, and a second verification result is generated based on the calorific value prediction value and the preset threshold. It is understood that in the embodiments of this application, the coal quality feature vector can be input into a pre-trained coal calorific value prediction model. This model learns the nonlinear mapping relationship between the coal's industrial analysis and elemental composition characteristics (such as ash content, volatile matter, sulfur content, moisture content, and hydrocarbon content) and its lower heating value, and outputs the calorific value prediction value of the corresponding sample. Then, the prediction value is compared with the originally reported measured calorific value, the deviation between the two is calculated, and a preset reasonable error threshold is used to determine whether the deviation exceeds the physical or empirically permissible range. If the deviation exceeds the threshold, the coal quality data is determined to be suspicious in the calorific value dimension, and an anomaly mark is generated as a second verification result. Thus, through data-driven calorific value consistency verification, problems such as false reporting, input errors, or inaccurate testing can be effectively identified.

[0058] In this embodiment, the processing method of the coal calorific value prediction model includes: calculating the Euclidean distance between each coal quality feature vector and each cluster centroid, determining the nearest cluster subgroup based on the Euclidean distance; calling the regression model corresponding to the nearest cluster subgroup; inputting the coal quality feature vector of the sample to be predicted into the regression model, and the regression model outputting the predicted calorific value of the coal quality data. It is understood that the embodiments of this application can cluster historical coal quality data to form several subgroups with similar coal quality characteristics, and calculate the centroid of each subgroup. In the prediction stage, the Euclidean distance between the feature vector of the coal to be tested and each cluster centroid is calculated, and the cluster subgroup with the closest distance is selected to determine its coal type or combustion characteristic category. Then, the regression model (such as XGBoost, random forest or linear regression) specifically trained for the subgroup is called, the feature vector is input into the model, and the targeted calorific value prediction value is output. Through this cluster-guided group modeling strategy, the model can establish a more accurate composition-calorific value mapping relationship for different coal types (such as lignite, bituminous coal, anthracite), effectively improving the accuracy and generalization ability of calorific value prediction, and providing a reliable basis for subsequent calorific value consistency verification.

[0059] Specifically, during online prediction: first, the sample to be predicted is mapped to the nearest cluster centroid. (Minimum Euclidean distance), then call the RANSAC (Random Sample Consensus) regression model corresponding to the cluster to output the predicted heat value. Simultaneously, based on the distribution of RANSAC in-point residuals, a prediction confidence interval is given (e.g., based on the standard deviation of the in-point residuals). structure The interval, where, This represents the point estimation result output by the model for the input sample x. Standard deviation (The two-tailed quantiles of the standard normal distribution) and calculate the reported calorific value. Compared with the predicted value deviation .when When the record exceeds a preset threshold (e.g., exceeding the prediction range or the relative error threshold), it can be used as evidence of doubt in the relevance dimension and further combined with the results of the aforementioned integrated doubt detection module to enhance the robustness of the overall verification.

[0060] For new coal quality records, the model first maps their compositional characteristics to the closest cluster category, then uses the corresponding regression model to predict their lower heating value. The model outputs the predicted calorific value and its confidence interval, which can be compared with the measured calorific value provided by the sample. If the actual calorific value of a record deviates significantly from the model's prediction, the record can be marked as a potential anomaly. This two-stage model effectively utilizes the category structure information contained in historical data, improving the accuracy and generalization ability of calorific value prediction. Even in the absence of laboratory calorific value measurements, this model can provide reliable estimates, supporting carbon emission calculations.

[0061] In this embodiment, the training method for the coal calorific value prediction model includes: dividing coal quality data with similar compositional structures into multiple cluster subgroups; within each cluster subgroup, fitting a regression model between the received basis lower calorific value and coal quality composition indicators, wherein candidate models are generated by repeatedly randomly sampling multiple samples from the coal quality data sample set, calculating the residual value of each sample in the coal quality data sample set, determining the inliers based on the residual values ​​and a preset threshold to generate an inlier set, selecting the candidate model with the largest number of inliers, refitting the candidate model based on the inliers to obtain the final regression parameters, and updating the regression model based on the final regression parameters.

[0062] It is understood that the embodiments of this application can divide coal quality data with similar compositional structures into multiple subgroups through clustering to ensure that the coal characteristics within each subgroup are consistent. Within each subgroup, a prediction model between the received basis low calorific value and coal quality composition indicators is constructed using a robust regression-based strategy: multiple candidate regression models are generated through multiple random samplings, the residuals of each sample under the candidate models are calculated, and inliers that conform to physical laws are selected according to a preset residual threshold to form an inlier set; the candidate model with the most inliers is selected as the optimal basis, and the regression parameters are refitted with its inlier set, thereby effectively suppressing the interference of outliers or noisy data on the model; the final regression model more accurately reflects the intrinsic relationship between the coal quality composition and calorific value, significantly improving the robustness and accuracy of calorific value prediction, and providing a highly reliable reference benchmark for subsequent data verification.

[0063] Specifically, Phase 1 involves K-Means clustering: using coal quality feature vectors that have undergone standardized field naming and unit conversion with reference standards. Let the number of clusters be . The objective is to minimize the intra-cluster squared error. ,in For the sample The centroid of the cluster Let be the feature vector of the j-th sample in the i-th cluster. During training, the centroid can be initialized and the "assign-update" step can be iteratively executed until... convergence; The value of can be selected from the candidate set using the silhouette coefficient or the elbow method. Clustering aims to divide coal samples with similar compositional structures into several subgroups, thereby reducing regression bias caused by mixing different coal types.

[0064] Phase 2 is RANSAC Consistency Regression: After clustering, the random sampling consistency regression algorithm is used to fit the relationship between the received basis low calorific value and other component indicators on each subclass data. Within each cluster subgroup, for the target variable (Received lower heating value) and independent variable To establish a regression relationship between (dry basis ash, dry basis total sulfur, and total moisture), a linear model can be used. ,in, This is the intercept term (constant term, bias). This is the weighted sum of all features and their corresponding coefficients. This represents the random error term (noise, residuals). To combat the bias of outliers (which may correspond to spurious heat values ​​or data entry errors) on the regression, robust fitting is performed using random sampling consistency. (i) Let the minimum number of samples be... The minimum number of points required for the selected regression model (for linear regression, this can be taken as...). (Feature count +1), randomly sample from the current cluster. Candidate parameters are obtained by fitting samples. ; (ii) Calculate the residuals of the entire sample ,in, For the j-th sample, Let be the output value predicted by the current candidate model in the t-th iteration, and let be the threshold value determined by the interior points. Determine the set of interior points ;in, Let be the set of sample indices for all samples whose residuals are less than the threshold δ under the model fitted by the random sampling in round t.

[0065] (iii) Repeated iteration Next, select the number of interior points. The largest model is taken as the optimal consensus set, and the least squares fitting is re-performed using the points within this consensus set to obtain the final parameters. Number of iterations Can be based on success probability Proportion with interior points Approximate settings: To ensure that in a given At least one "all interior points" subsample is drawn, where p is the expected success probability, w is the estimated proportion of interior points, and m is the minimum number of samples required to fit the model.

[0066] In step S104, a verification report is generated based on the first and second verification results of the coal quality data.

[0067] It is understood that the embodiments of this application can integrate the first verification result output by the anomaly detection model and the second verification result generated by the calorific value prediction model, and use a logical fusion mechanism to make a multi-dimensional judgment on the credibility of coal quality data: if any verification result is marked as abnormal, a warning of the corresponding level is triggered; if both are abnormal, it is judged as high-confidence problematic data; if both are normal, it is considered a reliable record; on this basis, combined with information such as sample source, degree of deviation, and anomaly type, a verification report containing data status, anomaly cause, confidence level and processing suggestions is generated in a structured manner, realizing the upgrade from single indicator verification to comprehensive credibility assessment, and providing traceable, interpretable and operable quality control basis for carbon emission accounting, fuel management and data governance.

[0068] Specifically, such as Figure 2 As shown, this application includes three stages: data preparation, model training, and online verification. First, multi-source coal quality data is collected, and after format conversion, deduplication, completion, and cleaning, it is imported into the coal quality database. Next, in the model training stage, a large-scale historical sample is extracted from the database to train the aforementioned anomaly detection model integration and calorific value prediction model: the optimal anomaly detection sub-model parameters and each clustering regression model are obtained through offline training. Finally, in the online verification stage: when new coal quality data needs to be verified, the system performs the following steps: 1) Data preprocessing: standardize and complete the new data according to the same rules as the database; 2) Question mark detection: input the preprocessed data into the integrated anomaly detection module, each sub-model calculates the anomaly score in parallel, and the integrated decision mechanism outputs the question mark judgment result; 3) Calorific value verification: input the data into the calorific value prediction module, calculate the predicted calorific value, and compare it with the reported calorific value of the data to obtain the difference; 4) Result synthesis: the system summarizes the question mark detection results and calorific value deviation information, and generates a verification report, including whether each data point is questionable, the overall confidence level, the judgment status of each model, and information such as the measured value and predicted value of calorific value and the difference.

[0069] According to the coal quality data processing method proposed in this application, multi-source coal quality data is integrated and uniformly transformed into standardized coal quality feature vectors. First, an anomaly detection model, which consists of multiple sub-models working collaboratively, is used to assess the confidence level of each data point, generating a first verification result reflecting the overall rationality of the data. Simultaneously, the same feature vector is input into a specially constructed coal calorific value prediction model to obtain a calorific value prediction value based on the component correlation relationship. This prediction value is then combined with a preset deviation threshold to determine the credibility of the measured calorific value, forming a second verification result. Finally, the two verification results are merged to generate a comprehensive verification report. This achieves multi-dimensional intelligent verification of the logical consistency, physical rationality, and numerical reliability of coal quality data without the need for real labels, significantly improving the quality and credibility of the basic data on which carbon emission accounting depends.

[0070] Next, the coal quality data processing apparatus proposed according to the embodiments of this application is described with reference to the accompanying drawings.

[0071] Figure 3 This is a block diagram of a coal quality data processing device according to an embodiment of this application.

[0072] like Figure 3 As shown, the coal quality data processing device 10 includes: an acquisition module 100, a first processing module 200, a second processing module 300, and a generation module 400.

[0073] The system includes the following modules: an acquisition module 100 acquires coal quality data from different data sources and converts the coal quality data into coal quality feature vectors; a first processing module 200 inputs each coal quality feature vector into an anomaly detection model, which outputs a first verification result for the corresponding coal quality data. The anomaly detection model calculates the confidence level of the coal quality data collaboratively through at least one sub-model and generates the corresponding first verification result based on the confidence level; a second processing module 300 inputs the coal quality feature vectors into a coal calorific value prediction model, which outputs the predicted calorific value of the coal quality data and generates a second verification result based on the predicted calorific value and a preset threshold; and a generation module 400 generates a verification report based on the first and second verification results of the coal quality data.

[0074] Specifically, such as Figure 4 As shown, this application consists of a data source, a data processing module, a database, a verification module for questionable coal quality, a calorific value prediction module, a business backend, and a user frontend interface. The data source includes raw coal quality data provided by coal mines, power plants, etc., which is cleaned and integrated by the data processing module and then stored in a unified database. The verification module for questionable coal quality and the calorific value prediction module can be deployed as independent services, and the business backend calls them through interfaces to complete the verification calculations. The frontend interface provides data entry and result display functions, ultimately achieving automatic verification and visual presentation of coal quality data.

[0075] According to the coal quality data processing device proposed in this application, by integrating multi-source coal quality data and uniformly converting it into a standardized coal quality feature vector, the device first uses an anomaly detection model with multiple sub-models working collaboratively to assess the confidence level of each data point, generating a first verification result reflecting the overall rationality of the data. Simultaneously, the same feature vector is input into a specially constructed coal calorific value prediction model to obtain a calorific value prediction value based on the component correlation relationship, and the credibility of the measured calorific value is judged by combining it with a preset deviation threshold, forming a second verification result. Finally, the two verification results are merged to generate a comprehensive verification report, thereby achieving multi-dimensional intelligent verification of the logical consistency, physical rationality, and numerical reliability of coal quality data without the need for real labels, significantly improving the quality and credibility of the basic data on which carbon emission accounting depends.

[0076] Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: The memory 501, the processor 502, and the computer program stored on the memory 501 and capable of running on the processor 502.

[0077] When the processor 502 executes the program, it implements the coal quality data processing method provided in the above embodiments.

[0078] Furthermore, electronic devices also include: Communication interface 503 is used for communication between memory 501 and processor 502.

[0079] The memory 501 is used to store computer programs that can run on the processor 502.

[0080] Memory 501 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0081] If the memory 501, processor 502, and communication interface 503 are implemented independently, then the communication interface 503, memory 501, and processor 502 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 5The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0082] Optionally, in a specific implementation, if the memory 501, processor 502, and communication interface 503 are integrated on a single chip, then the memory 501, processor 502, and communication interface 503 can communicate with each other through an internal interface.

[0083] Processor 502 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0084] This application also provides a computer-readable storage medium storing a computer program or instructions thereon, which, when executed by a processor, implements the coal quality data processing method described above.

[0085] This application also provides a computer program product, including a computer program or instructions, which, when executed, implement the coal quality data processing method described above.

[0086] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0087] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0088] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0089] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or more of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0090] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

Claims

1. A method for processing coal quality data, characterized in that, Includes the following steps: Obtain coal quality data from different data sources and convert the coal quality data into coal quality feature vectors; Each coal quality feature vector is input into the anomaly detection model, and the anomaly detection model outputs the first verification result of the corresponding coal quality data. The anomaly detection model calculates the confidence level of the coal quality data collaboratively through at least one sub-model, and generates the corresponding first verification result based on the confidence level. The coal quality feature vector is input into the coal calorific value prediction model, the coal calorific value prediction model outputs the calorific value prediction value of the coal quality data, and a second verification result is generated based on the calorific value prediction value and a preset threshold. A verification report is generated based on the first verification result and the second verification result of the coal quality data.

2. The coal quality data processing method according to claim 1, characterized in that, Before converting the coal quality data into a coal quality feature vector, the process includes: Extract the sample information field and coal quality information field from each record in the coal quality data after field mapping and benchmark conversion; Generate a key-value sequence based on the sample information field and the coal quality information field; Calculate the corresponding hash value based on the string concatenated from the key-value sequence; Duplicate coal quality data are removed based on the hash value; The deduplicated coal quality data is converted into a coal quality feature vector.

3. The coal quality data processing method according to claim 1, characterized in that, The anomaly detection model includes: a density-based hierarchical clustering sub-model, an isolated forest sub-model, a single-class support vector machine, and a Gaussian mixture sub-model, wherein... The density-based hierarchical clustering sub-model is used to calculate the core distance and the inter-sample reach distance of each sample point based on the coal quality feature vector, construct a density hierarchy based on the core distance and the inter-sample reach distance, output noise labels and outlier scores, and generate a first anomaly score based on the noise labels and outlier scores. The isolated forest sub-model is used to generate a second anomaly score based on the average path length of a sample in the forest by constructing multiple random binary trees and calculating the average path length of the sample. The single-class support vector machine is used to map coal quality feature vectors to a high-dimensional space through a kernel function, use a decision function to determine the relationship between samples and boundaries, output boundary distance scores, and generate a third anomaly score based on the boundary distance scores. The Gaussian mixture sub-model is used to estimate the parameters of the Gaussian mixture model using the expectation-maximization algorithm to fit the global probability distribution of the samples, calculate the negative log-likelihood of the samples, and generate a fourth anomaly score based on the negative log-likelihood.

4. The coal quality data processing method according to claim 3, characterized in that, The processing method of the anomaly detection model includes: The first anomaly score, the second anomaly score, the third anomaly score, and the fourth anomaly score are normalized to generate corresponding confidence levels. Get the number of hard votes for each sub-model; The overall confidence level is calculated based on the confidence level and their respective weights, and the first verification result of the coal quality data is determined based on the overall confidence level and the number of hard votes.

5. The coal quality data processing method according to claim 1, characterized in that, The processing method of the coal calorific value prediction model includes: Calculate the Euclidean distance between each coal quality feature vector and each cluster centroid, and determine the nearest cluster subgroup based on the Euclidean distance; Call the regression model corresponding to the nearest cluster subgroup; The coal quality feature vector of the sample to be predicted is input into the regression model, and the regression model outputs the predicted calorific value of the coal quality data.

6. The coal quality data processing method according to claim 5, characterized in that, The training method for the coal calorific value prediction model includes: Coal quality data with similar compositional structures are divided into multiple cluster subgroups; Within each cluster subgroup, a regression model is fitted between the received basis lower calorific value and coal quality composition indicators. This involves generating candidate models by repeatedly randomly sampling multiple samples from the coal quality data sample set, calculating the residual value for each sample in the coal quality data sample set, determining inliers based on the residual value and a preset threshold to generate an inlier set, selecting the candidate model with the largest number of inliers, refitting based on the inliers of the candidate model to obtain the final regression parameters, and updating the regression model based on the final regression parameters.

7. A coal quality data processing device, characterized in that, include: The acquisition module is used to acquire coal quality data from different data sources and convert the coal quality data into coal quality feature vectors; The first processing module is used to input each coal quality feature vector into the anomaly detection model, and the anomaly detection model outputs the first verification result of the corresponding coal quality data. The anomaly detection model calculates the confidence level of the coal quality data collaboratively through at least one sub-model and generates the corresponding first verification result based on the confidence level. The second processing module is used to input the coal quality feature vector into the coal calorific value prediction model, the coal calorific value prediction model outputs the calorific value prediction value of the coal quality data, and generate a second verification result based on the calorific value prediction value and a preset threshold. The generation module is used to generate a verification report based on the first verification result and the second verification result of the coal quality data.

8. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the program to implement the coal quality data processing method as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they are used to implement the coal quality data processing method as described in any one of claims 1-6.

10. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed, they implement the coal quality data processing method as described in any one of claims 1-6.

Images