Intelligent cashier system transaction data classification and analysis method based on artificial intelligence

Abstract

The invention discloses an intelligent cashier system transaction data classification and analysis method based on artificial intelligence, and the method comprises the following steps: S1, carrying out the preprocessing of the attributes of transaction data, and forming a dynamic space-time oscillogram; s2, forming a dynamic transaction graph through a self-adaptive space-time curved surface generator; s3, capturing the implicit state of the dynamic transaction graph based on the implicit state self-generation reasoning network, and automatically generating an implicit state mode in the transaction data; s4, performing multi-level classification based on the implicit state fluctuation matrix, and automatically generating and adjusting a classification strategy in combination with a spatio-temporal fluctuation learning algorithm; s5, reasoning is carried out through a self-driven reasoning engine, and path optimization is carried out; s6, performing future state deduction based on a high-order implicit state, identifying potential abnormal fluctuation in real time, and performing deep analysis; and S7, generating a visual chart through a multi-layer waveform visualization technology. According to the method, intelligent transaction data classification, anomaly detection and prediction analysis are realized by utilizing multi-dimensional space-time mapping and an implicit state reasoning network.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence and big data analysis technology, and in particular to a method for classifying and analyzing transaction data of an intelligent cash register system based on artificial intelligence. Background Art

[0002] In the current field of intelligent cash register systems and big data analysis, the analysis and processing of transaction data is one of the most important links in various industries. Traditional cash register systems usually use static rule-based processing methods combined with predefined algorithm models to classify, count and analyze data. Although this method can meet basic operational needs, with the rapid growth of transaction data volume, the increasing complexity of data structure and the diversification of transaction patterns, existing technical means are stretched when faced with large amounts of complex data.

[0003] Most existing smart cashier systems are based on discrete transaction data processing methods, which means that the system can only record and analyze data at fixed times or event points. This discrete processing is difficult to capture dynamic spatiotemporal characteristics in the face of real-time changing transaction environments. For example, customers' purchasing behavior, payment habits, transaction amounts, etc. will show highly dynamic trends in different time periods, locations, and transaction modes, and existing technologies are difficult to capture these changes in real time, resulting in delayed and inaccurate analysis of transaction data.

[0004] Traditional transaction data classification mainly relies on rule-based systems or simple machine learning algorithms, usually based on historical data for pattern analysis and classification. However, with the intensification of market changes and the complexity of consumption patterns, single historical data analysis is no longer able to meet the needs of accurate classification in complex transaction scenarios. For example, during promotional activities, holidays or special marketing scenarios, transaction data often fluctuates violently, which may be a normal market response or may imply some abnormal behavior or potential risks. Traditional classification algorithms often cannot adjust the model in time to adapt to changes in the environment, and lack sufficient adaptability and intelligence.

[0005] In addition, most of the existing anomaly detection technologies are based on preset threshold judgments or rule matching. The drawback of this method is that it has a poor ability to capture unknown anomalies or potential hidden states. Once an abnormal pattern that has never appeared before is encountered, the system may misjudge or even miss the detection. Especially in the face of complex and multi-dimensional data environments, the efficiency and accuracy of anomaly detection will be significantly reduced. Traditional prediction models are mostly based on linear extrapolation of time series, which cannot accurately capture nonlinear trends and sudden changes in data, and it is difficult to achieve accurate prediction of the future state of transaction data.

[0006] The existing intelligent cashier system analysis framework also lacks sufficient feedback mechanisms. Usually, data processing and reasoning are one-way, and the system analyzes data according to pre-set rules or models, but the analysis results are not fully utilized to reversely optimize the system itself. This lack of closed-loop feedback makes it impossible for the system to achieve self-learning and adaptive adjustment, resulting in the system being unable to quickly adjust and optimize when facing complex changes in the transaction environment, thereby reducing the real-time and accuracy of the analysis.

[0007] In addition, the existing intelligent cashier system lacks the ability to mine deep implicit patterns. There are a large number of implicit patterns and complex spatiotemporal relationships in transaction data. This information cannot be directly expressed through explicit data, but it profoundly affects transaction behavior and data trends. Traditional data analysis methods mostly rely on explicit features for analysis, and fail to deeply mine the potential patterns and nonlinear changes in the data, so it is easy to miss the capture and analysis of key trends or abnormal behaviors.

[0008] Another significant shortcoming of existing technologies is the limited means of data visualization and decision support. Traditional cash register systems often only provide basic two-dimensional data charts, which are difficult to intuitively display the temporal and spatial fluctuations and dynamic changes of transaction data. This causes merchants or managers to face a large cognitive burden when analyzing data, and it is difficult to quickly extract key information from the data. At the same time, when providing decision support, existing systems are often based on static historical data analysis results, lack foresight, and are difficult to provide sufficient support for future operational decisions.

[0009] Based on the above-mentioned defects of the prior art, it can be seen that the existing intelligent cashier system has multiple limitations in transaction data processing and analysis, which are mainly reflected in: First, it is impossible to capture the dynamic changes of transaction data in real time; second, the classification and anomaly detection models lack sufficient adaptive capabilities and are difficult to cope with dynamic changes in complex transaction environments; third, the prediction model relies too much on linear analysis and lacks the ability to predict complex nonlinear data; fourth, the system lacks an effective closed-loop feedback mechanism and is difficult to achieve self-optimization and adjustment; fifth, the ability to mine and deeply analyze implicit patterns is insufficient, resulting in insufficient depth of understanding of the data; sixth, the visualization means and decision support are not intelligent enough to intuitively reflect the dynamic trends and implicit features in the data.

[0010] Therefore, how to provide a transaction data classification and analysis method for an intelligent cash register system based on artificial intelligence is an urgent problem that technicians in this field need to solve. Summary of the invention

[0011] One purpose of the present invention is to propose a method for classifying and analyzing transaction data of an intelligent cash register system based on artificial intelligence. The present invention realizes real-time classification, anomaly detection and future state deduction analysis of transaction data of an intelligent cash register system by adopting technologies such as multi-dimensional space-time vector mapping, adaptive space-time surface generator and implicit state self-generated reasoning network. The present invention can not only dynamically capture the spatiotemporal changes in transaction data, but also deeply explore the potential patterns in the data through multi-level implicit state analysis, and accurately predict future trends. Compared with traditional methods, the present invention has a high degree of real-time response capability, data processing accuracy and adaptability, especially in the face of complex and dynamic transaction environments, it shows stronger analysis and decision support capabilities.

[0012] The transaction data classification and analysis method of the intelligent cashier system based on artificial intelligence according to an embodiment of the present invention comprises the following steps:

[0013] S1. Preprocess the attributes of each transaction data, and use a multi-dimensional space-time vector mapping function to generate a continuous waveform from the discrete transaction data to form a dynamic space-time waveform diagram;

[0014] S2. Through the adaptive space-time surface generator, the generated dynamic space-time waveform graph is projected into the adaptive multi-dimensional space, and the surface space generation structure is dynamically adjusted to change in real time with the change of transaction mode, forming a dynamic transaction map;

[0015] S3, based on the implicit state self-generation reasoning network, the dynamic transaction graph after the spatiotemporal waveform mapping is captured in the implicit state, and the implicit state pattern in the transaction data is automatically generated through the adaptively generated internal reasoning structure;

[0016] S4. Perform multi-level classification of implicit state patterns in trading data based on implicit state volatility matrix, generate intelligent stratification strategy by combining spatiotemporal volatility learning algorithm, and automatically generate and adjust classification strategy according to changes in trading environment;

[0017] S5. Use the self-driving reasoning engine to reason about the classified transaction data, optimize the path based on real-time feedback information, and adjust the analysis results in real time in combination with the closed-loop feedback system;

[0018] S6. Based on the future state deduction of high-order implicit states, the future pattern of transaction data is predicted, and abnormal fluctuation detection is realized by combining spatiotemporal fluctuation data, and potential abnormal fluctuations are identified in real time and analyzed in depth;

[0019] S7. Generate visual charts through multi-layer waveform visualization technology to display the spatiotemporal waveform changes of transaction data, implicit state reasoning results and future transaction trends, and generate intelligent decision support suggestions based on reasoning and classification results.

[0020] Optionally, the S2 specifically includes:

[0021] S21. Project the generated dynamic space-time waveform graph into the adaptive multi-dimensional space through the adaptive space-time surface generator to generate the initial space-time surface C0(t,x):

[0022]

[0023] Among them, t represents the time dimension, x represents the transaction attribute dimension, T represents the time period, L represents the amplitude parameter of the transaction attribute, and k(t,x) represents the spatiotemporal coupling coefficient;

[0024] S22, mapping the initial space-time surface to a high-dimensional space through a dynamic projection matrix, wherein the projection adopts a dynamic adjustment mechanism, and the projection matrix is ​​adaptively adjusted in real time as time and transaction patterns change;

[0025] S23. After the initial space-time surface projection, adaptive adjustments are made according to changes in transaction data. The adjustment mechanism is based on the high-order dynamic response of time and transaction attributes, and is corrected through second-order partial derivatives and third-order partial derivatives:

[0026]

[0027] Among them, C α (t,x) represents the surface after adaptive adjustment, exp represents the exponential function;

[0028] S24. After the system monitors the transaction data in real time, the surface is further optimized through the feedback mechanism:

[0029]

[0030] Among them, C β (t,x) represents the surface after feedback optimization, and λ represents the feedback optimization coefficient;

[0031] S25, generating an adaptive dynamic space-time surface, wherein the dynamic space-time surface is used to reflect changes in transaction patterns in real time;

[0032] S26. Based on the adjusted and optimized dynamic space-time surface, a final dynamic transaction map is generated. The dynamic transaction map reflects the fluctuation and trend changes of transaction data in real time through multi-dimensional projection and adaptive adjustment.

[0033] Optionally, the S3 specifically includes:

[0034] S31. In the dynamic transaction graph after space-time waveform mapping, the initial implicit state is captured. The implicit state captures the multi-dimensional features based on the current transaction graph and generates the implicit state vector through complex nonlinear transformation:

[0035]

[0036] Where H0(t,x) represents the initial implicit state at time t and spatial dimension x, C(t,x) represents the input data of the dynamic transaction graph, σ represents the activation function, W1 and W2 represent the weight matrices of the first and second layers respectively, b1 and b2 represent the bias terms of the first and second layers respectively, α1 represents the time adjustment coefficient, γ and δ represent the adjustment coefficients, and exp represents the exponential function;

[0037] S32. In the implicit state self-generation reasoning network, a deeper implicit state is generated layer by layer. Each layer of the implicit state depends on the output of the previous layer, and new implicit features are generated progressively.

[0038] S33. Through the adaptive generation mechanism, the implicit state is adjusted according to the dynamic changes of transaction data. At each time step, the implicit state is optimized and adaptively updated according to the historical state and current transaction graph data:

[0039]

[0040] Among them, H new (t,x) represents the new hidden state generated adaptively, H prev (t,x) represents the implicit state of the previous moment; θ(t,x) represents the adaptive adjustment coefficient, which controls the balance between the historical implicit state and the current transaction graph, and the range is 0≤θ(t,x)≤1; W3 represents the weight matrix of the third layer, α2 represents the spatial adjustment coefficient, It represents the capture of the periodic characteristics of the spatial dimension x, T1 represents the spatial period, and exp represents the exponential function;

[0041] S34, through the progressive calculation of the multi-layer implicit state reasoning network, the complex features in the transaction data are captured, and a deeper implicit state is gradually generated;

[0042] S35. By real-time monitoring the difference between the input of transaction data and the system output, the generation and reasoning path of the implicit state are adjusted according to the feedback mechanism;

[0043] S36. Finally, an implicit pattern in the transaction data is generated.

[0044] Optionally, the S4 specifically includes:

[0045] S41. Based on the implicit state pattern generated by the implicit state self-generation reasoning network, the implicit state fluctuation matrix is ​​constructed. The construction of the matrix combines the high-order change rate of time and space, as well as the nonlinear response to the fluctuation of transaction data:

[0046]

[0047] Among them, M H (t,x) represents the implicit state fluctuation matrix, H(t,x) represents the implicit state at time t and spatial dimension x, β1 represents the spatial adjustment coefficient, and β2 represents the temporal adjustment coefficient;

[0048] S42. The transaction data is classified at multiple levels through the volatility characteristics in the implicit volatility matrix. The classification mechanism comprehensively considers the volatility characteristics at the current moment and the historical trend:

[0049]

[0050] Among them, C class (t,x) represents the classification result of transaction data in time and space dimensions, λ high (t,x) represents the high classification threshold, λ low (t,x) represents the low classification threshold, ΔH(t,x) represents the change amplitude of the hidden state, ∈1 and ∈2 represent the change amplitude threshold;

[0051] C class (t,x) = 2 means that the transaction data is in a highly volatile state, indicating that there may be abnormalities or emergencies;

[0052] C class (t,x)=1 means that the trading data is in a medium fluctuation state, which represents the normal trading pattern fluctuation;

[0053] C class (t,x)=0 means that the transaction data is in a low volatility or static state, which means that the transaction data is in a relatively stable state;

[0054] S43. Combined with the spatiotemporal fluctuation learning algorithm, the change pattern in the implicit state fluctuation matrix is ​​captured by performing multi-dimensional analysis on the temporal and spatial fluctuations of the transaction data. The spatiotemporal fluctuation learning algorithm performs weighted integration on the transaction data to identify the periodic characteristics and sudden anomalies in the fluctuations:

[0055]

[0056] Among them, F STFLA (t,x) represents the output feature of the spatiotemporal fluctuation learning algorithm, ζ represents the adjustment coefficient, M H (t',x') represents the implicit state fluctuation matrix in the time t' and space x' dimensions, T represents the time window, and L represents the spatial range;

[0057] S44, the fluctuation feature F extracted by the spatiotemporal fluctuation learning algorithm STFLA(t, x), generate an intelligent stratification strategy, which is based on the multi-dimensional characteristics of the implicit state volatility matrix, combines the volatility information in time and space dimensions, divides the transaction data into multiple levels, and adjusts the stratification strategy in real time according to changes in the transaction environment;

[0058] S45. According to the real-time monitored transaction data fluctuations, the classification strategy is adaptively adjusted, and the classification threshold is dynamically adjusted through the feedback information of the spatiotemporal fluctuation learning algorithm;

[0059] S46. Finally, the results of the implicit state fluctuation matrix and the spatiotemporal fluctuation learning algorithm are combined to output multi-level classification results.

[0060] Optionally, the S5 specifically includes:

[0061] S51, generating an initial reasoning path for transaction data through a self-driven reasoning engine, wherein the generation of the reasoning path is based on the classified transaction data C class (t,x) and the implicit state fluctuation matrix M H (t, x), the construction of the inference path depends on the spatiotemporal characteristics and historical changes of the current data, and the multi-dimensional changes of transaction data are captured through a complex nonlinear mapping function f(·):

[0062]

[0063] Among them, P init (t,x) represents the initial reasoning path, C class (t,x) represents the classification result of transaction data, M H (t,x) represents the implicit state fluctuation matrix, γ1 and γ2 represent the adjustment coefficients, Represents the gradient information of the classification result, Represents the gradient information of the fluctuation matrix;

[0064] S52. Through the feedback mechanism, the fluctuation of current transaction data is monitored in real time. The feedback information captures the sudden change of transaction trend by analyzing the local changes of transaction data in time and space:

[0065]

[0066] Among them, F feedback (t,x) represents the metric value of feedback information, ν1 represents the influence factor of adjustment time on the reasoning path, and ν2 represents the influence factor of adjustment space on the reasoning path;

[0067] S53. Based on the feedback information, the initial reasoning path is adaptively optimized. The path optimization mechanism considers the current feedback information and introduces high-order changes in time and space dimensions:

[0068]

[0069] Among them, P opt (t,x) represents the optimized reasoning path, exp represents the exponential function, μ1 and μ2 represent the adjustment coefficients;

[0070] S54, further adjusting the optimization path through a closed-loop feedback mechanism, wherein the closed-loop feedback mechanism is based on the current path and combined with dynamic changes of the historical reasoning path;

[0071] S55, the closed-loop feedback mechanism combines the difference between the reasoning path at the previous moment and the current optimization path to correct the reasoning result:

[0072]

[0073] Among them, R closed-loop (t,x) represents the reasoning path adjusted by closed-loop feedback, R prev (t,x) represents the previous reasoning result, η represents the feedback adjustment coefficient;

[0074] S56, finally output the optimized reasoning path P finial (t,x).

[0075] Optionally, the S6 specifically includes:

[0076] S61. The transaction data is deduced through a future state deduction model, wherein the future state deduction model adopts the spatiotemporal characteristics of a high-order implicit state and combines multiple derivatives and nonlinear combinations of spatiotemporal dimensions:

[0077]

[0078] Among them, F predict (t+Δt,x) represents the implicit state fluctuation matrix at future time t+Δt and position x, M H (t,x) represents the implicit state fluctuation matrix, exp represents the exponential function, β6 and δ4 represent the adjustment coefficients, and ξ represents the nonlinear combination coefficient;

[0079] S62. Analyze the dependencies in different time and space dimensions through the future state deduction results. The time-space dependency analysis introduces high-order derivatives in the space dimension to capture the local fluctuations of future trading patterns and identify trading hot spots and potential abnormal behaviors.

[0080] S63. The abnormal fluctuation detection model detects abnormal fluctuations by analyzing the difference between the future state deduction and the current state. The abnormal fluctuation detection model combines the multi-dimensional nonlinear change characteristics in time and space to perform complex abnormal pattern recognition:

[0081]

[0082] Among them, A detect (t+Δt,x) represents the abnormal fluctuation value at the future time t+Δt and position x, τ and represents the nonlinear adjustment coefficient;

[0083] S64, real-time monitoring of abnormal fluctuations in transaction data, timely feedback of abnormal detection information and automatic adjustment of detection parameters by comparing current data with future deduction results;

[0084] S65. After identifying abnormal fluctuations, further combine the current trading pattern and future state deduction results to conduct in-depth analysis of the abnormal fluctuations, identify the source of the abnormal fluctuations by correlating historical data, and provide detailed analysis reports for decision support.

[0085] The beneficial effects of the present invention are:

[0086] First, the present invention uses multi-dimensional space-time vector mapping technology to transform traditional discrete transaction data into continuous space-time waveforms. This dynamic waveform mapping can capture the space-time change characteristics of transaction data in real time, effectively solving the limitations of the static and discrete processing mode of transaction data in the prior art. The system can accurately track the fluctuation trend of transaction data and provide more delicate and dynamic analysis results, especially when dealing with complex and rapidly changing transaction scenarios, and can timely capture the significant features in transaction data.

[0087] Secondly, by using the adaptive space-time surface generator, the present invention can project the dynamic space-time waveform graph into multidimensional space and generate a dynamic transaction map that adjusts with the change of transaction mode. This map not only reflects the current state of transaction data, but also ensures the real-time and accuracy of data analysis through the adaptive adjustment mechanism, so that the system has a strong self-regulation ability and can flexibly adapt to sudden changes in the transaction environment. Compared with the fixed classification and analysis framework in the prior art, the adaptive surface generator of the present invention significantly enhances the dynamic response capability of the system and lays a solid foundation for the in-depth analysis of transaction data.

[0088] The introduction of the implicit state self-generated reasoning network enables the present invention to effectively capture the implicit patterns and potential features in the transaction data. The network can not only extract conventional transaction patterns from explicit data, but also identify deep implicit states through a multi-layer progressive reasoning structure, breaking through the limitation of traditional data analysis that only relies on explicit features. Through the automatically generated implicit state reasoning model, the present invention can deeply mine the complex patterns in the data, so that the system can still make accurate and forward-looking analysis and judgment when facing unknown or complex transaction environments.

[0089] In addition, the present invention can identify periodic characteristics and sudden anomalies when analyzing the spatiotemporal fluctuations of transaction data through a spatiotemporal fluctuation learning algorithm. This algorithm can adaptively adjust the analysis path based on the fluctuation characteristics, ensuring that the system can respond quickly and make real-time adjustments when facing a complex trading environment. This algorithm greatly improves the adaptability of the system, so that the classification results can not only reflect the current trading situation, but also foresee future trend changes.

[0090] In terms of anomaly detection and future state deduction, the present invention realizes accurate prediction of future trends of transaction data based on the future state deduction model of high-order implicit states. Compared with the traditional linear time series prediction model, this model combines multiple derivatives and nonlinear combinations in the time and space dimensions, and can capture nonlinear changes and sudden trends in transaction data, making the prediction results more accurate and forward-looking. At the same time, through the abnormal fluctuation detection model, the present invention can monitor abnormal fluctuations in transaction data in real time and accurately identify possible risk events or abnormal behaviors. This capability effectively makes up for the shortcomings of existing anomaly detection technology that relies on rules or preset thresholds, so that the system has higher anomaly recognition accuracy, especially when facing unknown or complex transaction patterns, it can still efficiently detect anomalies.

[0091] Finally, the present invention significantly improves the intuitiveness of data display and the accuracy of decision support through multi-level visualization and intelligent decision support. The system can generate multi-dimensional visualization charts based on spatiotemporal waveforms and implicit state reasoning, dynamically display the changing trends and implicit patterns of transaction data, and enable users to intuitively understand the dynamic changes behind the data. At the same time, based on the reasoning and classification results, the system can provide users with forward-looking decision-making suggestions, helping merchants to respond to market changes in advance in operational decisions, thereby improving the level of intelligence in business operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0092] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0093] Figure 1 This is a flow chart of the transaction data classification and analysis method of the intelligent cash register system based on artificial intelligence proposed by the present invention;

[0094] Figure 2 A structural schematic diagram of a dynamic transaction graph generated by a space-time surface generator of the transaction data classification and analysis method of an intelligent cashier system based on artificial intelligence proposed in the present invention. DETAILED DESCRIPTION

[0095] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0096] refer to Figure 1 and Figure 2 , a transaction data classification and analysis method for an intelligent cash register system based on artificial intelligence includes the following steps:

[0097] S1. Preprocess the attributes of each transaction data, and use a multi-dimensional space-time vector mapping function to generate a continuous waveform from the discrete transaction data to form a dynamic space-time waveform diagram;

[0098] S2. Through the adaptive space-time surface generator, the generated dynamic space-time waveform graph is projected into the adaptive multi-dimensional space, and the surface space generation structure is dynamically adjusted to change in real time with the change of transaction mode, forming a dynamic transaction map;

[0099] S3, based on the implicit state self-generation reasoning network, the dynamic transaction graph after the spatiotemporal waveform mapping is captured in the implicit state, and the implicit state pattern in the transaction data is automatically generated through the adaptively generated internal reasoning structure;

[0100] S4. Perform multi-level classification of implicit state patterns in trading data based on implicit state volatility matrix, generate intelligent stratification strategy by combining spatiotemporal volatility learning algorithm, and automatically generate and adjust classification strategy according to changes in trading environment;

[0101] S5. Use the self-driving reasoning engine to reason about the classified transaction data, optimize the path based on real-time feedback information, and adjust the analysis results in real time in combination with the closed-loop feedback system;

[0102] S6. Based on the future state deduction of high-order implicit states, the future pattern of transaction data is predicted, and abnormal fluctuation detection is realized by combining spatiotemporal fluctuation data, and potential abnormal fluctuations are identified in real time and analyzed in depth;

[0103] S7. Generate visual charts through multi-layer waveform visualization technology to display the spatiotemporal waveform changes of transaction data, implicit state reasoning results and future transaction trends, and generate intelligent decision support suggestions based on reasoning and classification results.

[0104] In this implementation, S2 specifically includes:

[0105] S21. Project the generated dynamic space-time waveform graph into the adaptive multi-dimensional space through the adaptive space-time surface generator to generate the initial space-time surface C0(t,x):

[0106]

[0107] Among them, t represents the time dimension, x represents the transaction attribute dimension, T represents the time period, L represents the amplitude parameter of the transaction attribute, and k(t,x) represents the spatiotemporal coupling coefficient;

[0108] S22, mapping the initial space-time surface to a high-dimensional space through a dynamic projection matrix, wherein the projection adopts a dynamic adjustment mechanism, and the projection matrix is ​​adaptively adjusted in real time as time and transaction patterns change;

[0109] S23. After the initial space-time surface projection, adaptive adjustments are made according to changes in transaction data. The adjustment mechanism is based on the high-order dynamic response of time and transaction attributes, and is corrected through second-order partial derivatives and third-order partial derivatives:

[0110]

[0111] Among them, C α (t,x) represents the surface after adaptive adjustment, exp represents the exponential function;

[0112] S24. After the system monitors the transaction data in real time, the surface is further optimized through the feedback mechanism:

[0113]

[0114] Among them, C β (t,x) represents the surface after feedback optimization, and λ represents the feedback optimization coefficient;

[0115] S25, generating an adaptive dynamic space-time surface, wherein the dynamic space-time surface is used to reflect changes in transaction patterns in real time;

[0116] S26. Based on the adjusted and optimized dynamic space-time surface, a final dynamic transaction map is generated. The dynamic transaction map reflects the fluctuation and trend changes of transaction data in real time through multi-dimensional projection and adaptive adjustment.

[0117] In this implementation, S3 specifically includes:

[0118] S31. In the dynamic transaction graph after space-time waveform mapping, the initial implicit state is captured. The implicit state captures the multi-dimensional features based on the current transaction graph and generates the implicit state vector through complex nonlinear transformation:

[0119]

[0120] Where H0(t,x) represents the initial implicit state at time t and spatial dimension x, C(t,x) represents the input data of the dynamic transaction graph, σ represents the activation function, W1 and W2 represent the weight matrices of the first and second layers respectively, b1 and b2 represent the bias terms of the first and second layers respectively, α1 represents the time adjustment coefficient, γ and δ represent the adjustment coefficients, and exp represents the exponential function;

[0121] S32. In the implicit state self-generation reasoning network, a deeper implicit state is generated layer by layer. Each layer of the implicit state depends on the output of the previous layer, and new implicit features are generated progressively.

[0122] S33. Through the adaptive generation mechanism, the implicit state is adjusted according to the dynamic changes of transaction data. At each time step, the implicit state is optimized and adaptively updated according to the historical state and current transaction graph data:

[0123]

[0124] Among them, H new (t,x) represents the new hidden state generated adaptively, H prev (t,x) represents the implicit state of the previous moment; θ(t,x) represents the adaptive adjustment coefficient, which controls the balance between the historical implicit state and the current transaction graph, and the range is 0≤θ(t,x)≤1; W3 represents the weight matrix of the third layer, α2 represents the spatial adjustment coefficient, It represents the capture of the periodic characteristics of the spatial dimension x, T1 represents the spatial period, and exp represents the exponential function;

[0125] S34, through the progressive calculation of the multi-layer implicit state reasoning network, the complex features in the transaction data are captured, and a deeper implicit state is gradually generated;

[0126] S35. By real-time monitoring the difference between the input of transaction data and the system output, the generation and reasoning path of the implicit state are adjusted according to the feedback mechanism;

[0127] S36. Finally, an implicit pattern in the transaction data is generated.

[0128] In this implementation manner, the S4 specifically includes:

[0129] S41. Based on the implicit state pattern generated by the implicit state self-generation reasoning network, the implicit state fluctuation matrix is ​​constructed. The construction of the matrix combines the high-order change rate of time and space, as well as the nonlinear response to the fluctuation of transaction data:

[0130]

[0131] Among them, M H(t,x) represents the implicit state fluctuation matrix, H(t,x) represents the implicit state at time t and spatial dimension x, β1 represents the spatial adjustment coefficient, and β2 represents the temporal adjustment coefficient;

[0132] S42. The transaction data is classified at multiple levels through the volatility characteristics in the implicit volatility matrix. The classification mechanism comprehensively considers the volatility characteristics at the current moment and the historical trend:

[0133]

[0134] Among them, C class (t,x) represents the classification result of transaction data in time and space dimensions, λ high (t,x) represents the high classification threshold, λ low (t,x) represents the low classification threshold, ΔH(t,x) represents the change amplitude of the hidden state, ∈1 and ∈2 represent the change amplitude threshold;

[0135] C class (t,x) = 2 means that the transaction data is in a highly volatile state, indicating that there may be abnormalities or emergencies;

[0136] C class (t,x)=1 means that the trading data is in a medium fluctuation state, which represents the normal trading pattern fluctuation;

[0137] C class (t,x)=0 means that the transaction data is in a low volatility or static state, which means that the transaction data is in a relatively stable state;

[0138] S43. Combined with the spatiotemporal fluctuation learning algorithm, the change pattern in the implicit state fluctuation matrix is ​​captured by performing multi-dimensional analysis on the temporal and spatial fluctuations of the transaction data. The spatiotemporal fluctuation learning algorithm performs weighted integration on the transaction data to identify the periodic characteristics and sudden anomalies in the fluctuations:

[0139]

[0140] Among them, F STFLA (t,x) represents the output feature of the spatiotemporal fluctuation learning algorithm, ζ represents the adjustment coefficient, M H (t',x') represents the implicit state fluctuation matrix in the time t' and space x' dimensions, T represents the time window, and L represents the spatial range;

[0141] S44, the fluctuation feature F extracted by the spatiotemporal fluctuation learning algorithm STFLA(t, x), generate an intelligent stratification strategy, which is based on the multi-dimensional characteristics of the implicit state volatility matrix, combines the volatility information in time and space dimensions, divides the transaction data into multiple levels, and adjusts the stratification strategy in real time according to changes in the transaction environment;

[0142] S45. According to the real-time monitored transaction data fluctuations, the classification strategy is adaptively adjusted, and the classification threshold is dynamically adjusted through the feedback information of the spatiotemporal fluctuation learning algorithm;

[0143] S46. Finally, the results of the implicit state fluctuation matrix and the spatiotemporal fluctuation learning algorithm are combined to output multi-level classification results.

[0144] In this implementation manner, S5 specifically includes:

[0145] S51, generating an initial reasoning path for transaction data through a self-driven reasoning engine, wherein the generation of the reasoning path is based on the classified transaction data C class (t,x) and the implicit state fluctuation matrix M H (t, x), the construction of the inference path depends on the spatiotemporal characteristics and historical changes of the current data, and the multi-dimensional changes of transaction data are captured through a complex nonlinear mapping function f(·):

[0146]

[0147] Among them, P init (t,x) represents the initial reasoning path, C class (t,x) represents the classification result of transaction data, M H (t,x) represents the implicit state fluctuation matrix, γ1 and γ2 represent the adjustment coefficients, Represents the gradient information of the classification result, Represents the gradient information of the fluctuation matrix;

[0148] S52. Through the feedback mechanism, the fluctuation of current transaction data is monitored in real time. The feedback information captures the sudden change of transaction trend by analyzing the local changes of transaction data in time and space:

[0149]

[0150] Among them, F feedback (t,x) represents the metric value of feedback information, ν1 represents the influence factor of adjustment time on the reasoning path, and ν2 represents the influence factor of adjustment space on the reasoning path;

[0151] S53. Based on the feedback information, the initial reasoning path is adaptively optimized. The path optimization mechanism considers the current feedback information and introduces high-order changes in time and space dimensions:

[0152]

[0153] Among them, P opt (t,x) represents the optimized reasoning path, exp represents the exponential function, μ1 and μ2 represent the adjustment coefficients;

[0154] S54, further adjusting the optimization path through a closed-loop feedback mechanism, wherein the closed-loop feedback mechanism is based on the current path and combined with dynamic changes of the historical reasoning path;

[0155] S55, the closed-loop feedback mechanism combines the difference between the reasoning path at the previous moment and the current optimization path to correct the reasoning result:

[0156]

[0157] Among them, R closed-loop (t,x) represents the reasoning path adjusted by closed-loop feedback, R prev (t,x) represents the previous reasoning result, η represents the feedback adjustment coefficient;

[0158] S56, finally output the optimized reasoning path P finial (t,x).

[0159] In this implementation manner, S6 specifically includes:

[0160] S61. The transaction data is deduced through a future state deduction model, wherein the future state deduction model adopts the spatiotemporal characteristics of a high-order implicit state and combines multiple derivatives and nonlinear combinations of spatiotemporal dimensions:

[0161]

[0162] Among them, F predict (t+Δt,x) represents the implicit state fluctuation matrix at future time t+Δt and position x, M H (t,x) represents the implicit state fluctuation matrix, exp represents the exponential function, β6 and δ4 represent the adjustment coefficients, and ξ represents the nonlinear combination coefficient;

[0163] S62. Analyze the dependencies in different time and space dimensions through the future state deduction results. The time-space dependency analysis introduces high-order derivatives in the space dimension to capture the local fluctuations of future trading patterns and identify trading hot spots and potential abnormal behaviors.

[0164] S63. The abnormal fluctuation detection model detects abnormal fluctuations by analyzing the difference between the future state deduction and the current state. The abnormal fluctuation detection model combines the multi-dimensional nonlinear change characteristics in time and space to perform complex abnormal pattern recognition:

[0165]

[0166] Among them, A detect (t+Δt,x) represents the abnormal fluctuation value at the future time t+Δt and position x, τ and represents the nonlinear adjustment coefficient;

[0167] S64, real-time monitoring of abnormal fluctuations in transaction data, timely feedback of abnormal detection information and automatic adjustment of detection parameters by comparing current data with future deduction results;

[0168] S65. After identifying abnormal fluctuations, further combine the current trading pattern and future state deduction results to conduct in-depth analysis of the abnormal fluctuations, identify the source of the abnormal fluctuations by correlating historical data, and provide detailed analysis reports for decision support.

[0169] Embodiment 1:

[0170] In order to verify the feasibility of the present invention in implementation, the present invention is applied to a large shopping mall. During peak hours, such as during the Double Eleven promotion, the number of customers increases sharply and transactions are frequent. At this time, the transaction data shows great volatility and complexity, including a sharp increase in sales of different commodity categories in different time periods, sudden changes in customer shopping behavior, and inventory shortages. Traditional cash register systems and data analysis methods are unable to capture and respond to these instantaneous changes in real time and accurately, resulting in a lag in decision-making for merchants, and they are unable to adjust product supply, optimize promotion strategies, or quickly respond to abnormal transaction behaviors in a timely manner.

[0171] The transaction data classification and analysis method of the intelligent cash register system based on artificial intelligence of the present invention is designed for the transaction data fluctuation and complex dynamic environment in the above-mentioned scenario. It dynamically processes transaction data through multi-dimensional space-time vector mapping, adaptive space-time surface generation, implicit state self-generation reasoning network and space-time fluctuation learning algorithm, realizes intelligent classification, anomaly detection, future trend prediction and other functions, effectively helping merchants to cope with the ever-changing market environment and make forward-looking decisions.

[0172] During the Double Eleven promotion on November 11, 2023, the present invention was applied to the cash register system of the large shopping mall to monitor the transaction data of the entire promotion day in real time. The mall includes more than 5,000 kinds of goods in multiple different areas, with a transaction volume of up to 500 transactions per minute, especially during peak hours at noon and in the evening, with a transaction volume of up to 800 transactions per minute. Due to limited processing power, traditional transaction analysis systems are difficult to capture the temporal and spatial fluctuations and anomalies in these transaction data in real time, resulting in inventory shortages that cannot be replenished in time, and the mall's promotion strategy cannot be adjusted according to real-time data. The present invention solves this problem through the following steps:

[0173] First, after each transaction is completed in the mall, the system converts the data of each transaction (such as transaction time, commodity category, amount, etc.) into a continuous space-time waveform through multi-dimensional space-time vector mapping, and generates a dynamic space-time waveform graph in real time. This mapping process can capture the time and space characteristics of transaction data, allowing the system to intuitively understand the fluctuations of transaction data over time and place. During the peak promotion period, the space-time waveform graph shows that the sales volume of certain commodity categories suddenly increased. For example, the sales volume of a certain brand of electronic products in a certain hour soared to 10 times the daily average sales volume. Through this waveform graph, merchants can quickly perceive the sales changes of different commodities and adjust inventory and promotion efforts in a timely manner.

[0174] Secondly, the system uses an adaptive space-time surface generator to project the generated dynamic space-time waveform graph into an adaptive multi-dimensional space to dynamically generate a transaction map. Through the adaptive adjustment mechanism, the system can make real-time adjustments as transaction data fluctuates. The results show that when the sales of electronic products fluctuate significantly, the system automatically makes adjustment suggestions for the inventory of related products and pushes them to managers to ensure that popular products are not out of stock. During this process, the system's response time is less than 3 seconds, which is significantly faster than the lag time of traditional systems (about 30 minutes). This real-time adjustment capability ensures that merchants can respond quickly to transaction data, thereby avoiding decision-making errors caused by delayed response in traditional systems.

[0175] In addition, the system also conducts in-depth analysis of transaction data through implicit self-generated reasoning networks to capture implicit patterns in transaction data. For example, the system found that the sales patterns of a certain type of home appliances in different regions were significantly different. The sales of some products were low at the beginning of the promotion, but suddenly surged towards the end of the event. After the implicit reasoning network identified this trend, it automatically suggested that merchants increase publicity for these products in the middle of the promotion to guide consumer behavior in advance. Through this deep reasoning analysis, the system can predict future sales trends and provide merchants with forward-looking decision support.

[0176] In terms of abnormal fluctuation detection, the present invention realizes real-time detection of potential abnormal transactions through a high-order implicit future state deduction model. During the peak promotion period, the system found that certain payment behaviors (such as large payments or frequent returns) showed abnormal fluctuations. These abnormal transaction behaviors were accurately captured by the deduction model and triggered an immediate alarm through the system's feedback mechanism, reminding managers to further review such transactions.

[0177] In addition, the spatiotemporal fluctuation learning algorithm of the present invention also automatically generates intelligent stratification strategies by identifying periodic features of large-scale transaction data. For example, after analyzing the transaction data of the past 48 hours, the system identified that there are obvious peak and trough cycles in commodity sales, and hierarchically managed commodity categories based on these cycle characteristics, suggesting that merchants increase promotion efforts for popular commodities in specific time periods, or adjust the prices of inventory commodities during trough periods.

[0178] Finally, through the multi-layer waveform visualization chart generated by the present invention, merchants can intuitively view the spatiotemporal fluctuations of transaction data and the implicit state reasoning results. Data shows that on the day of Double Eleven, through the real-time visualization charts generated by the system, merchants can fully grasp the sales hotspots, inventory fluctuations, abnormal transactions and other situations, and effectively assist decision-making management. After the Double Eleven promotion ended, the system predicted the next inventory demand and market trends through the future state deduction model, helping merchants to adjust their business strategies in advance and prepare for subsequent market activities.

[0179] Table 1 Performance comparison between the smart cashier system based on artificial intelligence and the traditional system

[0180]

[0181]

[0182] According to the data analysis in Table 1 above, we can clearly see the significant advantages of the system of the present invention in multiple key indicators. First, in terms of system response time, the response time of the traditional system is about 30 minutes, which obviously has a large lag in the fast-changing trading environment and cannot meet the needs of real-time data processing. The response time of the system of the present invention is shortened to 3 seconds, which means that the system can process transaction data in an instant and make real-time responses, greatly improving the timeliness of decision-making and operational efficiency.

[0183] In terms of abnormal transaction detection, the traditional system can only detect less than 100 abnormal transactions, and the accuracy rate is less than 80%, which makes it impossible to identify many potential risk transactions in time, increasing the risk of merchants being defrauded. In contrast, the system of the present invention detected 400 abnormal transactions, and the accuracy rate reached more than 90%. The system not only increases the number of detections, but also improves the accuracy of detection, effectively reducing transaction risks.

[0184] The inventory turnover rate has also been significantly improved. The inventory turnover rate of the traditional system is only 5%, while the system of the present invention has reached 15%. This change shows that the system of the present invention helps merchants to better optimize inventory configuration through accurate transaction data analysis and inventory management suggestions, improves the speed of commodity circulation, and reduces the risk of out-of-stock and inventory backlog.

[0185] In the comparison of stock-out rates, the performance of the system of the present invention is particularly outstanding. The stock-out rate of the traditional system is 15%, while the system of the present invention reduces this ratio to 8%. This means that the system of the present invention can timely identify the demand fluctuations of popular commodities and quickly feedback replenishment suggestions, helping merchants to effectively prevent the occurrence of stock-outs and improving customers' shopping experience.

[0186] In terms of sales growth rate, the traditional system has a growth rate of 8%, while the sales growth rate of the system of the present invention has reached 20%. Through more accurate market demand forecasting and more efficient inventory management, the system of the present invention has created higher income for merchants and directly brought about improved operational efficiency.

[0187] Another important comparison is the abnormal transaction detection time. The traditional system takes an average of 10 minutes to detect abnormal transactions, while the system of the present invention only takes 1 minute, which greatly shortens the detection time. This allows merchants to respond to abnormal transactions more quickly and reduce potential losses.

[0188] The daily transaction processing volume is also a highlight of the system of the present invention. The traditional system can process 30,000 transactions per day, while the processing capacity of the system of the present invention reaches 60,000 transactions. This doubled processing capacity shows that the system of the present invention can maintain stable and efficient operation when facing large-scale transaction data, and will not have processing delays or data backlogs due to excessive transaction volume.

[0189] Finally, in terms of processing efficiency during promotions, the processing efficiency of the traditional system dropped by 30% during peak trading hours, while the efficiency of the system of the present invention dropped by only 5%. This significant improvement means that even when the transaction volume increases dramatically, the system of the present invention can still maintain efficient operation and ensure smooth operation of merchants during the peak promotion period.

[0190] In summary, through the comparative analysis of the above tables, it can be clearly seen that the intelligent cash register system of the present invention is significantly superior to the traditional system in multiple dimensions such as transaction data processing, anomaly detection, inventory management and sales forecasting. This substantial performance improvement not only improves the speed and accuracy of data processing, but also enhances the system's adaptability in complex and dynamic transaction environments, providing merchants with more intelligent and efficient operational support.

[0191] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for classifying and analyzing transaction data of an intelligent cash register system based on artificial intelligence, characterized in that: The steps include: S1. Preprocess the attributes of each transaction data, and use a multi-dimensional space-time vector mapping function to generate a continuous waveform from the discrete transaction data to form a dynamic space-time waveform diagram; S2. Through the adaptive space-time surface generator, the generated dynamic space-time waveform graph is projected into the adaptive multi-dimensional space, and the surface space generation structure is dynamically adjusted to change in real time with the change of transaction mode, forming a dynamic transaction map; S3, based on the implicit state self-generation reasoning network, the implicit state of the dynamic transaction graph after the spatiotemporal waveform mapping is captured, and the implicit state pattern in the transaction data is automatically generated through the adaptively generated internal reasoning structure; S4. Perform multi-level classification of implicit state patterns in trading data based on implicit state volatility matrix, generate intelligent stratification strategy by combining spatiotemporal volatility learning algorithm, and automatically generate and adjust classification strategy according to changes in trading environment; S5. Use the self-driving reasoning engine to reason about the classified transaction data, optimize the path based on real-time feedback information, and adjust the analysis results in real time in combination with the closed-loop feedback system; S6. Based on the future state deduction of high-order implicit states, the future pattern of transaction data is predicted, and abnormal fluctuation detection is realized by combining spatiotemporal fluctuation data, and potential abnormal fluctuations are identified in real time and analyzed in depth; S7. Generate visual charts through multi-layer waveform visualization technology to display the spatiotemporal waveform changes of transaction data, implicit state reasoning results and future transaction trends, and generate intelligent decision support suggestions based on reasoning and classification results.

2. The method for classifying and analyzing transaction data of an intelligent cash register system based on artificial intelligence according to claim 1 is characterized in that: The S2 specifically includes: S21. Project the generated dynamic space-time waveform graph into the adaptive multi-dimensional space through the adaptive space-time surface generator to generate the initial space-time surface C0(t,x): Where t represents the time dimension, x represents the transaction attribute dimension, T represents the time period, L represents the amplitude parameter of the transaction attribute, and k(t,x) represents the spatiotemporal coupling coefficient; S22, mapping the initial space-time surface to a high-dimensional space through a dynamic projection matrix, wherein the projection adopts a dynamic adjustment mechanism, and the projection matrix is ​​adaptively adjusted in real time as time and transaction patterns change; S23. After the initial space-time surface projection, adaptive adjustments are made according to changes in transaction data. The adjustment mechanism is based on the high-order dynamic response of time and transaction attributes, and is corrected through second-order partial derivatives and third-order partial derivatives: Among them, C α (t,x) represents the surface after adaptive adjustment, exp represents the exponential function; S24. After the system monitors the transaction data in real time, the surface is further optimized through the feedback mechanism: Among them, C β (t,x) represents the surface after feedback optimization, and λ represents the feedback optimization coefficient; S25, generating an adaptive dynamic space-time surface, wherein the dynamic space-time surface is used to reflect changes in transaction patterns in real time; S26. Based on the adjusted and optimized dynamic space-time surface, a final dynamic transaction map is generated. The dynamic transaction map reflects the fluctuation and trend changes of transaction data in real time through multi-dimensional projection and adaptive adjustment.

3. The method for classifying and analyzing transaction data of an intelligent cash register system based on artificial intelligence according to claim 1 is characterized in that: The S3 specifically includes: S31. In the dynamic transaction graph after space-time waveform mapping, the initial implicit state is captured. The implicit state captures the multi-dimensional features based on the current transaction graph and generates the implicit state vector through complex nonlinear transformation: Where H0(t,x) represents the initial implicit state at time t and spatial dimension x, C(t,x) represents the input data of the dynamic transaction graph, σ represents the activation function, W1 and W2 represent the weight matrices of the first and second layers respectively, b1 and b2 represent the bias terms of the first and second layers respectively, α1 represents the time adjustment coefficient, γ and δ represent the adjustment coefficients, and exp represents the exponential function; S32. In the implicit state self-generation reasoning network, a deeper implicit state is generated layer by layer. Each layer of the implicit state depends on the output of the previous layer, and new implicit features are generated progressively. S33. Through the adaptive generation mechanism, the implicit state is adjusted according to the dynamic changes of transaction data. At each time step, the implicit state is optimized and adaptively updated according to the historical state and current transaction graph data: Among them, H new (t,x) represents the new hidden state generated adaptively, H prev (t,x) represents the implicit state of the previous moment; θ(t,x) represents the adaptive adjustment coefficient, which controls the balance between the historical implicit state and the current transaction graph, and the range is 0≤θ(t,x)≤1; W3 represents the weight matrix of the third layer, α2 represents the spatial adjustment coefficient, It represents the capture of the periodic characteristics of the spatial dimension x, T1 represents the spatial period, and exp represents the exponential function; S34, through the progressive calculation of the multi-layer implicit state reasoning network, the complex features in the transaction data are captured, and a deeper implicit state is gradually generated; S35. By real-time monitoring the difference between the input of transaction data and the system output, the generation and reasoning path of the implicit state are adjusted according to the feedback mechanism; S36. Finally, an implicit pattern in the transaction data is generated.

4. The transaction data classification and analysis method of the intelligent cash register system based on artificial intelligence according to claim 1 is characterized in that: The S4 specifically includes: S41. Based on the implicit state pattern generated by the implicit state self-generation reasoning network, the implicit state fluctuation matrix is ​​constructed. The construction of the matrix combines the high-order change rate of time and space, as well as the nonlinear response to the fluctuation of transaction data: Among them, M H (t,x) represents the implicit state fluctuation matrix, H(t,x) represents the implicit state at time t and spatial dimension x, β1 represents the spatial adjustment coefficient, and β2 represents the temporal adjustment coefficient; S42. The transaction data is classified at multiple levels through the volatility characteristics in the implicit volatility matrix. The classification mechanism comprehensively considers the volatility characteristics at the current moment and the historical trend: Among them, C class (t,x) represents the classification result of transaction data in time and space dimensions, λ high (t,x) represents the high classification threshold, λ low (t,x) represents the low classification threshold, ΔH(t,x) represents the change amplitude of the hidden state, ∈1 and ∈2 represent the change amplitude threshold; C class (t,x) = 2 means that the transaction data is in a highly volatile state, indicating that there may be abnormalities or emergencies; C class (t,x)=1 means that the trading data is in a medium fluctuation state, which represents the normal trading pattern fluctuation; C class (t,x)=0 means that the transaction data is in a low volatility or static state, which means that the transaction data is in a relatively stable state; S43. Combined with the spatiotemporal fluctuation learning algorithm, the change pattern in the implicit state fluctuation matrix is ​​captured by performing multi-dimensional analysis on the temporal and spatial fluctuations of the transaction data. The spatiotemporal fluctuation learning algorithm performs weighted integration on the transaction data to identify the periodic characteristics and sudden anomalies in the fluctuations: Among them, F STFLA (t,x) represents the output feature of the spatiotemporal fluctuation learning algorithm, ζ represents the adjustment coefficient, M H (t',x') represents the implicit state fluctuation matrix in the time t' and space x' dimensions, T represents the time window, and L represents the spatial range; S44, the fluctuation feature F extracted by the spatiotemporal fluctuation learning algorithm STFLA (t, x), generate an intelligent stratification strategy, which is based on the multi-dimensional characteristics of the implicit state volatility matrix, combines the volatility information in time and space dimensions, divides the transaction data into multiple levels, and adjusts the stratification strategy in real time according to changes in the transaction environment; S45. According to the real-time monitored transaction data fluctuations, the classification strategy is adaptively adjusted, and the classification threshold is dynamically adjusted through the feedback information of the spatiotemporal fluctuation learning algorithm; S46. Finally, the results of the implicit state fluctuation matrix and the spatiotemporal fluctuation learning algorithm are combined to output multi-level classification results.

5. The method for classifying and analyzing transaction data of an intelligent cash register system based on artificial intelligence according to claim 1, characterized in that: The S5 specifically includes: S51, generating an initial reasoning path for transaction data through a self-driven reasoning engine, wherein the generation of the reasoning path is based on the classified transaction data C class (t,x) and the implicit state fluctuation matrix M H (t, x), the construction of the inference path depends on the spatiotemporal characteristics and historical changes of the current data, and the multi-dimensional changes of transaction data are captured through a complex nonlinear mapping function f(·): Among them, P init (t,x) represents the initial reasoning path, C class (t,x) represents the classification result of transaction data, M H (t,x) represents the implicit state fluctuation matrix, γ1 and γ2 represent the adjustment coefficients, Represents the gradient information of the classification result, Represents the gradient information of the fluctuation matrix; S52. Through the feedback mechanism, the fluctuation of current transaction data is monitored in real time. The feedback information captures the sudden change of transaction trend by analyzing the local changes of transaction data in time and space: Among them, F feedback (t,x) represents the metric value of feedback information, ν1 represents the influence factor of adjustment time on the reasoning path, and ν2 represents the influence factor of adjustment space on the reasoning path; S53. Based on the feedback information, the initial reasoning path is adaptively optimized. The path optimization mechanism considers the current feedback information and introduces high-order changes in time and space dimensions: Among them, P opt (t,x) represents the optimized reasoning path, exp represents the exponential function, μ1 and μ2 represent the adjustment coefficients; S54, further adjusting the optimization path through a closed-loop feedback mechanism, wherein the closed-loop feedback mechanism is based on the current path and combined with dynamic changes of the historical reasoning path; S55, the closed-loop feedback mechanism combines the difference between the reasoning path at the previous moment and the current optimization path to correct the reasoning result: Among them, R closed-loop (t,x) represents the reasoning path adjusted by closed-loop feedback, R prev (t,x) represents the previous reasoning result, η represents the feedback adjustment coefficient; S56, finally output the optimized reasoning path P finial (t,x).

6. The method for classifying and analyzing transaction data of an intelligent cash register system based on artificial intelligence according to claim 1, characterized in that: The S6 specifically includes: S61. The transaction data is deduced through a future state deduction model, wherein the future state deduction model adopts the spatiotemporal characteristics of a high-order implicit state and combines multiple derivatives and nonlinear combinations of spatiotemporal dimensions: Among them, F predict (t+Δt,x) represents the implicit state fluctuation matrix at future time t+Δt and position x, M H (t,x) represents the implicit state fluctuation matrix, exp represents the exponential function, β6 and δ4 represent the adjustment coefficients, and ξ represents the nonlinear combination coefficient; S62. Analyze the dependencies in different time and space dimensions through the future state deduction results. The time-space dependency analysis introduces high-order derivatives in the space dimension to capture the local fluctuations of future trading patterns and identify trading hot spots and potential abnormal behaviors. S63. The abnormal fluctuation detection model detects abnormal fluctuations by analyzing the difference between the future state deduction and the current state. The abnormal fluctuation detection model combines the multi-dimensional nonlinear change characteristics in time and space to perform complex abnormal pattern recognition: Among them, A detect (t+Δt,x) represents the abnormal fluctuation value at the future time t+Δt and position x, τ and represents the nonlinear adjustment coefficient; S64, real-time monitoring of abnormal fluctuations in transaction data, timely feedback of abnormal detection information and automatic adjustment of detection parameters by comparing current data with future deduction results; S65. After identifying abnormal fluctuations, further combine the current trading pattern and future state deduction results to conduct in-depth analysis of the abnormal fluctuations, identify the source of the abnormal fluctuations by correlating historical data, and provide detailed analysis reports for decision support.

Images