Intelligent pharmacy service evaluation method and system

Abstract

The invention discloses an intelligent pharmacy service evaluation method and system, and relates to the technical field of service evaluation. The data acquisition module acquires full-scene data of the pharmacy by using Internet of Things equipment; the data processing module realizes edge-cloud collaboration and cleans and analyzes data; the evaluation index construction module establishes a four-dimensional evaluation system; the intelligent evaluation module calculates scores through an improved algorithm and visually displays the scores; the optimization decision module generates a strategy by using reinforcement learning based on an evaluation result, and realizes closed-loop management in combination with a block chain; the system also has extended functions of cold chain monitoring, patient interaction and the like, and realizes intelligent evaluation and optimization of pharmacy services. According to the invention, the whole-process intelligent management of the intelligent pharmacy service is realized; through multi-modal data acquisition and intelligent analysis, the inventory accuracy, the service efficiency and the drug management level are improved; the service quality is accurately evaluated, and the complaint rate of patients is reduced; the closed-loop optimization decision reduces the drug loss, improves the satisfaction of the patient, and assists the pharmacy service to be efficiently upgraded.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of service evaluation, in particular to a smart pharmacy service evaluation method and system. BACKGROUND

[0002] With the advancement of medical informatization, smart pharmacy, as a key link to improve the efficiency and quality of medical services, has attracted widespread attention. Traditional pharmacy management relies on manual operation, and there are many problems in drug storage, deployment, distribution and patient service. For example, drug inventory relies on manual checking, which is low in efficiency and prone to errors, and often results in discrepancies between accounts and reality; pharmacists are limited by experience and energy in the process of prescription review and medication guidance, making it difficult to ensure the consistency of service quality; long waiting time for patients to take medicine and slow response to consultation result in low satisfaction. In addition, there is a lack of automated means in key links such as temperature and humidity monitoring of cold chain drugs and early warning management of near-expiry drugs, and there is a risk of drug loss.

[0003] The existing pharmacy informatization system introduces some digital technology, but still has obvious limitations. For example, some systems only realize electronic recording of drug inventory, and lack real-time tracking of drug expiration date, location and flow status; some intelligent shelves can monitor inventory quantity, but cannot identify the standardization of drug placement, affecting the efficiency of medicine taking. In terms of service quality evaluation, traditional methods are mostly based on manual sampling, patient questionnaire and other methods, and the data collection is not comprehensive and the analysis is lagging behind, making it difficult to find potential problems in the service process and providing targeted optimization strategies.

[0004] In terms of patient service, traditional pharmacies lack tracking and intervention of patient medication behavior. Pharmacists cannot obtain patient historical medication data and provide personalized medication guidance; patient complaint feedback channels are single and problem handling process is lengthy, resulting in low service improvement efficiency. At the same time, there are barriers between pharmacy and other systems of hospital (such as HIS, LIS) in data interaction, and information sharing is not timely, affecting the overall medical service collaboration. Therefore, an integrated multi-source data perception, intelligent analysis and closed-loop optimization smart pharmacy service evaluation system is needed to solve the problems of extensive management, lagging evaluation and low efficiency in the prior art. SUMMARY

[0005] The present application proposes a smart pharmacy service evaluation method and system to solve the problems mentioned in the prior art.

[0006] In order to achieve the above purpose, the present application adopts the following technical scheme: a smart pharmacy service evaluation method and system, comprising: Data-aware acquisition module: Deploy a high-frequency RFID and machine vision integrated medicine shelf. RFID tags achieve drug positioning, and a binocular vision Basler acA2500-14gm camera monitors drug placement norms in real time through the YOLOv8 algorithm. The window service terminal integrates a 3D structured light camera and a facial expression recognition module, combined with directional microphone array beamforming technology to collect service interaction data. The self-service dispensing machine is equipped with a pressure-sensitive matrix and a weight compensation algorithm to check dispensing errors. Edge-cloud collaborative processing module: Use NVIDIA Jetson AGX Orin edge computing devices to pre-process collected data in real time, use wavelet threshold denoising algorithms to process sensor data, and extract video frames using the VGG16 network. Establish a data hierarchical transmission mechanism to upload compressed non-real-time data to the cloud, and push abnormal data to the management terminal through the MQTT protocol. Dynamic weight evaluation index construction module: Construct a four-dimensional evaluation system, introduce a time decay factor a to dynamically adjust the index weight, and the formula is where is the adjusted weight, Wi is the original weight, and t is the number of index failure cycles. Add a medication adherence guidance index to evaluate the effectiveness of pharmacist guidance by analyzing the difference between patient historical medication intervals and medical orders. Establish a drug safety risk matrix and dynamically calculate the risk coefficient based on the incidence of adverse drug reactions and recall frequency data. Evaluation module: Use the analytic hierarchy process (IAHP) to correct subjective weights using the entropy weight method and construct a consistency check optimization model. Introduce gray correlation analysis (GRA) to compare the correlation of different pharmacy service indicators with industry benchmarks. Develop a three-dimensional visual evaluation system to render a pharmacy service digital twin using the Unity engine, display service quality distribution in each area in the form of a heat map, and support VR panoramic viewing and interactive analysis.

[0007] Further, it further comprises: Closed-loop optimization decision module: Construct a service optimization strategy network based on deep reinforcement learning (DRQN) algorithm, and use patient waiting time and drug loss rate as reward functions. Develop a process simulation subsystem to simulate different optimization schemes through discrete event simulation. Establish a blockchain storage mechanism to record optimization task execution on the chain for storage and support full-process audit and traceability. The data acquisition module is equipped with a wearable monitoring unit, and pharmacists wear a bracelet PPG+ECG sensor to monitor work load and identify fatigue status through electromyographic signal analysis (EMG). The data processing module integrates a federated learning framework to train a service quality prediction model jointly with each hospital while protecting data privacy. The evaluation module develops an ethics review subsystem to automatically detect risks in medication guidance based on natural language processing (BERT).

[0008] Further, the evaluation index construction module introduces patient health portrait analysis, combines electronic medical record (EMR) data and drug taking records, and predicts drug taking demand through a graph neural network (GNN); the optimization decision module deploys a digital twin driven dynamic scheduling system, adjusts pharmacist post configuration according to real-time service data, and optimizes drug replenishment paths using a genetic algorithm; and the system integration module supports HL7 FHIR and DICOM protocol conversion, and realizes settlement linkage with a medical insurance cost control system.

[0009] Further, the edge-cloud collaborative processing module uses edge caching technology to locally store high-frequency access data; and data transmission uses a homomorphic encryption algorithm CKKS scheme to realize data analysis in a ciphertext state and protect data privacy and security.

[0010] Further, the three-dimensional visualization system of the evaluation module supports real-time picture remote consultation, and experts can label service problems in real time through the device; and the evaluation report generation uses natural language generation technology (NLG) to automatically generate a structured document including optimization suggestions.

[0011] Further, the blockchain storage of the closed-loop optimization decision module uses a consortium chain architecture Hyperledger Fabric, each block includes service data hash value, operation log, and timestamp, and supports automatic contract execution for task acceptance.

[0012] Further, the system further comprises: A patient interaction module: a voice assistant is developed to support dialect recognition; a drug knowledge graph is established to provide personalized drug taking reminders for patients through knowledge reasoning; and a federated recommendation algorithm FedRec is used to realize cross-institutional drug taking experience sharing.

[0013] A method of applying the intelligent pharmacy service evaluation system, comprising: A data perception collection step: a drug shelf scans an RFID tag every 5 minutes to update inventory, machine vision detects drug expiration labels in real time, a window service terminal collects video at a frame rate of 60 fps, a microphone array separates multiple voices using a blind source separation technology (BSS), and a self-service drug taking machine judges drug taking actions through pressure sensing and performs secondary verification based on weight changes; A hierarchical processing and transmission step: an edge device performs anomaly detection on sensor data and behavior recognition on video; and data is pushed in real time through a WebSockets protocol, and regular data is packaged and uploaded to cloud distributed storage on an hourly basis; A dynamic index calculation step: service efficiency weights are dynamically adjusted according to pharmacist workloads; satisfaction is quantified by analyzing patient expression changes; and drug demand trends are predicted using an LSTM, and inventory turnover rate warning thresholds are calculated; Comprehensive evaluation step: the index weight is determined by IAHP-entropy weight method, and the service score is calculated by combining gray correlation degree; the service process is simulated by digital twin system to identify bottleneck link; the visual report including risk thermal map and improvement priority matrix is generated; Closed-loop optimization execution step: the DRQN algorithm generates an optimization strategy, which is pushed to the management system after being verified by process simulation; the blockchain records the task execution process, and the contract automatically verifies the optimization effect; the evaluation model is iteratively updated according to the evaluation feedback.

[0014] Further, the data processing step includes jointly training a drug demand prediction model using a federated learning framework, and model parameter updates use a secure aggregation protocol SecureAggregation; low-quality video data is enhanced by a generative adversarial network GAN to improve behavior recognition accuracy.

[0015] Further, it also includes: Patient personalized service optimization step: generate early warning prompts for patient medication conflict risks through knowledge graph reasoning; push cross-institutional medication experience to patients based on federated recommendation algorithm; use ARKit technology to realize three-dimensional demonstration of drug use and improve patient medication compliance.

[0016] Compared with the existing technology, the beneficial effects of the present application are: In terms of data collection, a multi-modal sensor network realizes real-time sensing of drug whole-process state, pharmacist service behavior and patient feedback, and the inventory checking accuracy is improved from 89.2% to 99.6% through the fusion of RFID and machine vision, greatly reducing human error.

[0017] In terms of service efficiency optimization, edge-cloud collaborative processing and intelligent scheduling algorithm improve prescription processing efficiency by 73.3%, and the average waiting time of patients is shortened from 22 minutes to 10 minutes. Through dynamic weight evaluation index and intelligent multi-dimensional evaluation module, the system can accurately quantify service quality, pharmacist operation compliance rate is improved from 78% to 95%, and medication guidance completeness rate is improved to 98%, ensuring medication safety.

[0018] In terms of drug management, cold chain monitoring, near expiration warning and intelligent replenishment strategy reduce drug loss rate by 79.2%, and inventory turnover rate is significantly improved. Patient satisfaction is accurately evaluated and real-time feedback is realized through micro-expression recognition, sentiment analysis and other technologies, complaint rate is reduced by 83.3%, and comprehensive satisfaction is improved from 76 points to 92 points.

[0019] In addition, the closed-loop optimization decision module generates an optimal strategy based on reinforcement learning and process simulation, and combines blockchain storage to ensure that the optimization process is traceable, making the allocation of pharmacy resources more reasonable and reducing the workload of pharmacists by 20.5%. The system breaks the information island through data interconnection and intercommunication, provides a comprehensive and scientific pharmacy management solution for hospitals, and promotes the upgrading of intelligent medical services. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 A schematic block diagram of a smart pharmacy service evaluation system according to the present application is shown in the figure. Figure 2 A schematic block diagram of a smart pharmacy service evaluation method according to the present application is shown in the figure. Figure 3 A schematic diagram of the trend of drug management indicators in a smart pharmacy service evaluation method according to the present application is shown in the figure. Figure 4 A schematic diagram of service quality evaluation in a smart pharmacy service evaluation method according to the present application is shown in the figure. Figure 5 A schematic diagram of the effect of an optimization strategy in a smart pharmacy service evaluation method according to the present application is shown in the figure. DETAILED DESCRIPTION

[0021] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the present application.

[0022] In the description of the present application, it should be understood that the terms "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation of the present application.

[0023] In addition, the terms "first", "second", etc. are used only for the purpose of description and do not imply or imply relative importance or imply the number of the technical features indicated. Therefore, the features defined as "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "a plurality of" is two or more, unless otherwise specifically limited. In addition, the terms "mounting", "connection", "connection" should be broadly understood, for example, it can be fixedly connected, or it can be detachably connected, or integrally connected; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium, or it can be connected inside two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances, and the present application will be further described in detail below with reference to the accompanying drawings.

[0024] Reference Figures 1 to 5 : Embodiments of a smart pharmacy service evaluation method and system 1. Multi-modal data perception and acquisition module This module uses a distributed heterogeneous sensor network to realize full-scene data acquisition. The intelligent medicine shelf is deployed with an ultra-high frequency RFID reader (ThingMagic Mercury 6e), and the antenna array is designed with circular polarization. Time division multiplexing (TDMA) algorithm is used to realize multi-tag anti-collision, and the single inventory time is ≤30 seconds, and the reading accuracy is 99.8%. The binocular vision system calibrates the internal and external parameters through the ChArUco checkerboard calibration board, and uses SIFT feature matching combined with RANSAC algorithm to realize medicine posture recognition. When the detected medicine is placed at an angle of more than 15°, a six-degree-of-freedom mechanical arm (UR5e) is triggered to automatically adjust. The end effector of the mechanical arm is equipped with a tactile sensor (Pressure Profile Systems I-Scan), and the contact force is controlled within the range of 0.5-1.2N, ensuring non-destructive operation of the medicine.

[0025] The window service terminal integrated 3D structured light camera (Intel RealSense D455) acquires face depth information through time-of-flight (ToF) technology, and realizes micro-expression recognition using MTCNN+ResNet network, which can capture 12 typical expressions such as frowning and smiling (recognition accuracy ≥93%). The directional microphone array (Matrix Creator) uses the MVDR beamforming algorithm and performs noise suppression through spectral subtraction, and the speech recognition accuracy is still 97% in a noisy environment (signal-to-noise ratio ≥15dB). The pressure sensing matrix of the self-service dispensing machine is composed of 4096 piezoresistive sensors (Interlink FSR402), with a resolution of 2.5mm×2.5mm, and a weight compensation model is trained through a BP neural network to eliminate the influence of drug placement position deviation on weight detection, achieving high-precision verification of ±0.3g.

[0026] 2. Edge-cloud collaborative processing module The edge computing device deploys NVIDIA Jetson AGX Orin, and uses Docker containerization technology to realize service decoupling. The resource allocation strategy is: 4-core CPU+8GB RAM for deep learning inference service, and 2-core CPU+4GB RAM for data preprocessing service. The sensor data preprocessing uses sliding average filtering (window size=11) combined with wavelet threshold denoising (db4 wavelet basis, 4 decomposition layers), and the denoising effect is evaluated by signal-to-noise ratio (SNR), with an improvement of ≥12dB. The video key frame extraction uses a content-based frame selection algorithm, which determines the key frame by calculating the structural similarity index (SSIM<0.9) between adjacent frames, with a compression ratio of 1:80 while maintaining ≥95% of the semantic information.

[0027] Data hierarchical transmission uses Kafka message queue to realize traffic peak clipping, with 3 partitions and 2 replica factors. Critical abnormal data (such as drug expiration warning) is transmitted through dedicated network (SLA guarantee delay <20ms), and MQTT protocol (QoS2 level) is used to ensure reliable delivery. The cloud uses Elasticsearch to build search indexes, with 5 shards and 1 replica, supporting millisecond-level data retrieval. The Hive data warehouse stores data by topic partitioning, uses ORC file format (compression ratio 1:3.5), and optimizes query efficiency through partition pruning and predicate pushdown, reducing query response time by 60%.

[0028] 3. Dynamic weight evaluation index construction module The specific indicators of the four-dimensional evaluation system are as follows: Service efficiency: average prescription processing time (target value ≤3 minutes), average patient waiting time (target value ≤15 minutes), window utilization rate during peak period (target value ≥85%) Service quality: Pharmacist operation compliance rate (based on SOP process identification), medication guidance completeness rate (keyword coverage rate), prescription review accuracy rate (compared with HIS system). Drug management: Inventory accuracy (RFID inventory vs. system records), timeliness of handling near-expiry drugs (percentage of drugs with an expiry date <3 months), and temperature compliance rate of cold chain drugs (percentage in the 2-8℃ range). Patient satisfaction was evaluated using affective analysis scores, micro-expression satisfaction scores, and complaint resolution rates. A time decay factor α of 0.05 was used, meaning the indicator weights decreased by 5% per month. Medication adherence guidance indicators were calculated by comparing the patient's actual medication interval with the prescribed interval, using the following formula: (C represents compliance score,) This refers to the actual interval between medication dispensing. (For doctor's prescription intervals). The drug safety risk matrix is ​​updated in real time based on the FDA Adverse Drug Reaction Database (FAERS), and the risk coefficient is calculated using the following formula: (R is the risk coefficient, A is the number of adverse reaction reports per year, S is the severity score (1-10), and P is the frequency of drug use). High-risk drugs (R>5) are automatically given 1.5 times more weight in the evaluation.

[0029] 4. Intelligent Multidimensional Evaluation Module Implementation steps of the improved Analytic Hierarchy Process (IAHP): Construct a judgment matrix A = (aij)n×n, where aij represents the relative importance of indicators i and j (1-9 scale method). The judgment matrix is ​​optimized using the least squares method: ,in The weight of index i; Calculate the objective weights using the entropy weight method: ,in ( (where i is the weight of the i-th indicator in the j-th sample). Overall weighting of subjective and objective factors: (λ is set to 0.6, Grey relational analysis (with subjective weights) was used to calculate the correlation between pharmacy service indicators and industry benchmarks, with a resolution coefficient ρ of 0.5. The formula is as follows: ( As a reference sequence, (For comparison sequence). The 3D visualization system uses voxelization technology to render the pharmacy space, with a voxel resolution of 5cm×5cm×5cm. The heatmap uses the HSV color model to display the service quality distribution (red indicates high-risk areas, H∈[0,60], S=1, V=1). VR interaction supports gesture operations (such as zooming and rotating) and voice queries (such as "query the service efficiency of window 2"), with a response latency of ≤100ms.

[0030] 5. Closed-loop optimization decision module Deep reinforcement learning adopts DRQN algorithm, state space S contains 12 dimensions (such as current number of patients in queue, number of pharmacists, drug inventory level, time slice, weather condition, special event, etc.), action space A contains 8 strategies (such as increasing window, adjusting scheduling, prioritizing special patients, drug replenishment reminder, environment adjustment, service process optimization, personnel training, equipment maintenance). The reward function is designed as: wherein is the average waiting time (minutes), is the operating cost (yuan / hour), is the satisfaction score (0-100), is the process improvement index (percentage of year-on-year improvement).

[0031] The process simulation subsystem adopts discrete event modeling, the time step is set to 1 minute, and the simulation duration is 8 hours / time. The simulation parameters include: patient arrival rate (Poisson distribution λ=15 people / hour), service time (normal distribution μ=4 minutes, σ=1 minute), drug replenishment time (uniform distribution U[20,40] minutes). The blockchain adopts Hyperledger Fabric v2.4, the channel is configured with 3 endorsement nodes and 2 ordering nodes, the block size is set to 2MB, and the transaction confirmation time is ≤3 seconds. The smart contract is developed in Go language, which realizes the automatic acceptance logic of the task: when the optimized patient waiting time is reduced by ≥15% and meets the standard for 3 consecutive days, the reward is automatically triggered, and the reward amount calculation formula is (B is the reward amount, k is the coefficient 0.5 yuan / minute, is the time reduction amount, N is the number of benefited patients).

[0032] 6. System-based service evaluation method implementation steps Data-aware collection step: The RFID reader of the intelligent medicine cabinet deploys a 5-minute cycle, using frequency hopping technology (FHSS) to traverse the 902-928MHz frequency band, and can read more than 200 tags at a time. Combined with dynamic Q value anti-collision algorithm, it ensures that the inventory data update delay is less than 10 seconds. The machine vision system uses YOLOv8 algorithm to detect the expiration date label of the medicine, and corrects the label image through perspective transformation. After the fuzzy text is enhanced by super-resolution reconstruction technology (SRGAN), the OCR recognition accuracy reaches 98.7%. The window service terminal camera captures video at a frame rate of 60fps, and uses a motion target detection algorithm based on optical flow method (Farneback algorithm) to track the action trajectory of pharmacists and patients in real time. The microphone array removes environmental noise through a deep learning noise reduction model (Conv-TasNet) and converts voice data into text format. The pressure sensing matrix and weight sensor data of the self-service medicine dispensing machine are synchronized through timestamps, and when the difference between the two changes exceeds ±0.8g, a secondary verification process is triggered, and the mechanical arm is used to grab the medicine for position calibration and reweighing.

[0033] Hierarchical processing transmission step: The edge computing device processes sensor data in real time, and uses the 3σ principle to detect outliers. When the medicine inventory is less than 80% of the safety threshold, an early warning message is generated immediately. Video data first extracts human key points through OpenPose, and only uploads key frames containing valid actions (such as taking medicine and guiding gestures), with a compression ratio of 1:100. Regular data is packaged by hour, and then compressed again using the Zstandard algorithm (compression level 15), and uploaded to the cloud through the HTTP / 3 protocol; Key data (such as device fault alarm) is transmitted in binary format through the WebSocket protocol to ensure that it reaches the management terminal within 500ms. Cloud data storage uses a hot and cold separation strategy, storing high-frequency access data within 1 month in SSD cache, and migrating historical data exceeding 6 months to HDD storage, reducing storage costs by 35%.

[0034] Dynamic indicator calculation step: In the service efficiency indicator, the prescription processing time is calculated by analyzing the HIS system prescription generation time and window call timestamp. The peak period is defined as the period when the daily prescription volume is >300. The window utilization rate is determined according to the ratio of the pharmacist's on-duty time to the actual service time. In terms of service quality indicators, pharmacist operation compliance is compared with 128 pre-set SOP rules (such as drug checking process, medication guidance dialogue), and the F1 value is used to evaluate the compliance rate. The completeness rate of medication guidance is extracted by NLP technology to extract key information (such as dosage, contraindications, side effects) in the guidance voice, and matched with the standard knowledge base for calculation. In the drug management indicator, the inventory accuracy rate is calculated by the difference rate between the RFID inventory data and the system account. The cold chain drug temperature compliance rate is calculated based on the temperature and humidity data collected every 10 minutes. When the temperature exceeds the range of 2-8℃ and the duration is >15 minutes, it is marked as an abnormality.

[0035] Intelligent comprehensive evaluation step: When using the improved analytic hierarchy process (IAHP), the subjective judgment matrix is obtained through expert questionnaire, and the weight vector is solved using Python's numpy library. After consistency check (CI<0.1), the objective weight calculated by entropy weight method is fused. In gray correlation analysis, industry benchmark data is used as reference sequence to calculate the correlation degree of each pharmacy indicator. Indicators with correlation degree >0.8 are considered as advantage items, and indicators with correlation degree <0.6 are considered as improvement priorities. The three-dimensional visualization system renders the pharmacy digital twin through WebGL technology. Users can zoom in and out, drag and pan to view details. The color mapping of the heat map uses a linear segmented function to map the service score (0-100) to the HSV color domain (H∈[0,120], S=1, V=1). The evaluation report automatic generation module uses natural language generation technology (NLG) to generate a document containing specific problem descriptions and improvement suggestions based on indicator scores and abnormal analysis results. It supports export to PDF and Word formats.

[0036] The closed-loop optimization execution step: the deep reinforcement learning (DRQN) model is trained in the training phase, with 100,000 steps as a training cycle, using an experience replay mechanism (ReplayBuffer size of 500,000) to reduce data correlation, learning rate set to 0.001 and using exponential decay (decay rate 0.99). The generated optimization strategy needs to be verified by the process simulation subsystem, and the simulation is set to 1000 times of Monte Carlo simulation, calculating the average return, standard deviation and other indicators of the strategy, and selecting the strategy with a confidence interval of more than 95% to push to the pharmacy management system. Task orders are stored through blockchain, and each operation record contains the operator's digital signature, timestamp, and data hash value before and after the operation, and the Merkle tree structure is used to achieve fast verification. The optimization effect evaluation adopts A / B testing, and the pharmacy implementing the optimization strategy is compared with the control group for 2 weeks. When the key indicators (such as patient waiting time, satisfaction) are significantly improved (p<0.05), the strategy is formally adopted and the evaluation model parameters are updated.

[0037] 7. System implementation data effect

[0038] Data interpretation: The improvement of inventory accuracy is due to the fusion technology of RFID and machine vision, and the system can automatically identify the location and state of medicines, reducing human error. The significant improvement in prescription processing efficiency is due to the real-time preprocessing of data by edge computing and the optimization of intelligent scheduling algorithms. The patient complaint rate decreased by 83.3%, mainly due to the comprehensive monitoring and immediate feedback mechanism of service quality, such as micro-expression recognition technology to help pharmacists adjust communication methods in time. The drug expiration loss rate decreased by nearly 80%, which is the result of the synergistic effect of dynamic inventory management and near-expiration warning system. The improvement of pharmacists' work load index is due to the intelligent scheduling system and automated operation process, so that pharmacists can devote more energy to professional services. The shortening of system response time relies on the edge-cloud collaborative architecture and hierarchical data processing strategy, greatly improving user experience. The patient satisfaction score increased by 21.1%, which is a comprehensive improvement of service efficiency, quality and individualization. These data fully verify the significant advantages of the system in improving the efficiency, service quality and resource management of the pharmacy.

[0039] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto, and any skilled person in the art can make equivalent substitutions or changes within the technical scope disclosed by the present application according to the technical solution and inventive concept of the present application, which should be covered within the protection scope of the present application.

Claims

1. A method and system for evaluating smart pharmacy services, characterized in that, include: Data sensing and acquisition module: Deploys a medicine shelf that integrates UHF RFID and machine vision. RFID tags enable medicine location, and a binocular vision Baslerac A2500-14gm camera monitors the standardization of medicine placement in real time using the YOLOv8 algorithm; The window service terminal integrates a 3D structured light camera and a facial expression recognition module, combined with directional microphone array beamforming technology to collect service interaction data; The self-service medicine dispensing machine is equipped with a pressure sensing matrix and weight compensation algorithm to verify medicine dispensing errors; Edge-cloud collaborative processing module: It adopts NVIDIA Jetson AGX Orin edge computing device to perform real-time preprocessing of the collected data, uses wavelet threshold denoising algorithm to process sensor data, and extracts video frame count through VGG16 network; it establishes a data hierarchical transmission mechanism to compress the original video non-real-time data and upload it to the cloud, and pushes abnormal data to the management terminal through MQTT protocol; Dynamic weighted evaluation index construction module: Constructs a four-dimensional evaluation system, introducing a time decay factor α to dynamically adjust index weights, the formula is as follows: ,in The adjusted weights are represented by Wi, where Wi is the original weight and t is the number of failure cycles of the indicator. A new medication adherence guidance indicator is added, which assesses the effectiveness of pharmacist guidance by analyzing the differences between patients' historical medication intervals and medical orders. A drug safety risk matrix is ​​established, and the risk coefficient is dynamically calculated by combining data on adverse drug reaction rates and recall frequencies. Evaluation Module: The Analytic Hierarchy Process (IAHP) is used, and subjective weights are corrected by the entropy weight method to construct an optimized model for consistency testing; Grey Relational Analysis (GRA) is introduced to compare the correlation between different pharmacy service indicators and industry benchmarks. Develop a 3D visualization evaluation system that renders a digital twin of pharmacy services using the Unity engine, displays the distribution of service quality in each area in the form of a heat map, and supports VR panoramic viewing and interactive analysis.

2. The smart pharmacy service evaluation method and system according to claim 1, characterized in that, Also includes: Closed-loop optimization decision module: A service optimization strategy network is constructed based on the deep reinforcement learning DRQN ​​algorithm, and the reward functions are patient waiting time and drug loss rate; Develop a process simulation subsystem to simulate different optimization schemes through discrete events; establish a blockchain evidence storage mechanism to store the execution records of optimization tasks on the blockchain, supporting full-process auditing and traceability; The data acquisition module is equipped with a wearable monitoring unit. Pharmacists wear wristbands with PPG+ECG sensors to monitor their workload and identify fatigue status by analyzing EMG signals. The data processing module integrates a federated learning framework to train service quality prediction models in conjunction with various hospitals while protecting data privacy. The evaluation module developed an ethics review subsystem that automatically detects risks in medication guidance based on BERT natural language processing.

3. The smart pharmacy service evaluation method and system according to claim 1, characterized in that, It also includes an assessment indicator construction module that introduces patient health profile analysis, combines electronic medical record (EMR) data with medication records, and uses graph neural networks (GNN) to predict medication needs; an optimization decision module that deploys a digital twin-driven dynamic scheduling system, adjusts pharmacist job configurations based on real-time service data, and uses genetic algorithms to optimize drug replenishment paths; and a system integration module that supports HL7FHIR and DICOM protocol conversion to achieve settlement linkage with the medical insurance cost control system.

4. The smart pharmacy service evaluation system according to claim 1, characterized in that, The edge-cloud collaborative processing module uses edge caching technology to store frequently accessed data locally; data transmission uses the homomorphic encryption algorithm CKKS to achieve data analysis in encrypted form, ensuring data privacy and security.

5. The smart pharmacy service evaluation system according to claim 1, characterized in that, The 3D visualization system of the assessment module supports real-time remote consultation, and experts can annotate service issues in real time through the device; the assessment report is generated using natural language generation technology (NLG), which automatically generates a structured document including optimization suggestions.

6. The smart pharmacy service evaluation system according to claim 1, characterized in that, The blockchain evidence storage of the closed-loop optimization decision module adopts the Hyperledger Fabric consortium blockchain architecture. Each block includes a service data hash value, operation log, and timestamp, and supports automatic contract execution and task acceptance.

7. The smart pharmacy service evaluation system according to claim 1, characterized in that, Also includes: Patient interaction module: Develop a voice assistant that supports dialect recognition; Establish a medication knowledge graph and provide personalized medication reminders to patients through knowledge reasoning; The FedRec recommended algorithm is used to achieve cross-institutional sharing of drug use experience.

8. A method for applying the smart pharmacy service evaluation system according to any one of claims 1-7, characterized in that, include: Data sensing and collection steps: The medicine shelf is scanned with RFID tags every 5 minutes to update the inventory, and machine vision is used to detect expired medicine tags in real time; The window service terminal captures video at a frame rate of 60fps, and the microphone array uses blind source separation technology (BSS) to separate the individual spoken voices; the self-service medicine dispensing machine judges the medicine taking action through pressure sensing and performs secondary verification by combining weight changes. Hierarchical processing and transmission steps: Edge devices perform anomaly detection on sensor data and behavior recognition on video; data is pushed in real time via the WebSockets protocol, and regular data is packaged and uploaded to cloud distributed storage on an hourly basis; Dynamic indicator calculation steps: dynamically adjust the service efficiency weights based on the pharmacist's workload; Patient satisfaction is quantified by analyzing changes in patient facial expressions; LSTM is used to predict drug demand trends and calculate inventory turnover rate warning thresholds. Comprehensive evaluation steps: The IAHP-entropy weight method is used to determine the weight of indicators, and the service score is calculated by combining the grey relational analysis; the service process is simulated through a digital twin system to identify bottlenecks; and a visual report including a risk heatmap and an improvement priority matrix is ​​generated. Closed-loop optimization execution steps: The DRQN ​​algorithm generates an optimization strategy, which is then verified through process simulation and pushed to the management system; The blockchain records the task execution process, and the contract automatically verifies and optimizes the results; the evaluation model is iteratively updated based on evaluation feedback.

9. The smart pharmacy service evaluation method according to claim 8, characterized in that, The data processing steps include using a federated learning framework to jointly train a drug demand prediction model, with model parameter updates using the SecureAggregation protocol; and enhancing low-quality video data through Generative Adversarial Networks (GANs) to improve the accuracy of behavior recognition.

10. The smart pharmacy service evaluation method according to claim 8, characterized in that, Also includes: Steps to optimize personalized patient services: Generate early warning prompts by reasoning about the risk of medication conflicts in patients using knowledge graphs; Pushing cross-institutional medication experiences to patients based on a federated recommendation algorithm; ARKit technology can be used to create 3D demonstrations of drug use, thereby improving patient medication adherence.

Images