Dynamic personnel management method for hotel management

Abstract

The invention discloses a personnel dynamic management method for hotel management, and relates to the technical field of hotel personnel dynamic management, and the method comprises the steps: partitioning base stations based on hotel building information, calculating a structured index, and calculating the relative height according to the current floor air pressure to obtain discrete floor prior distribution; the base station signal intensity, the structured indexes and the like form feature vectors to be input into a neural network, and floor and corridor position unit posterior distribution is obtained; and fusing prior and posterior distribution to obtain a floor result, judging and generating a floor consistency label, and finally generating a dispatching instruction. According to the invention, the positioning precision and dispatching efficiency of hotel personnel can be improved.

Description

Technical Field

[0001] This invention relates to the field of dynamic personnel management technology in hotels, and in particular to a dynamic personnel management method for hotel management. Background Technology

[0002] With the rapid development of the hotel industry and the continuous expansion of hotel scale, large hotels often include dozens or even hundreds of floors of guest rooms, multiple dining areas, conference centers, fitness and entertainment areas, and complex back-of-house corridors. Daily operations require handling a large number of tasks such as room cleaning, linen delivery, customer response, and equipment inspection. The efficient execution of these tasks relies on real-time dynamic management of service personnel, especially the ability to quickly locate their current position and accurately assign tasks to employees closest to the target task area on the appropriate floor, thereby shortening response time and improving service efficiency and customer satisfaction. However, hotel building structures are significantly complex. For example, elevator shafts, as vertical passages traversing multiple floors, are prone to wireless signal leakage and abnormal propagation. Narrow corridors, due to space constraints, are prone to signal waveguide effects, resulting in significant differences in signal propagation intensity in different directions. Furthermore, the walls between adjacent floors in multi-story structures cannot completely block wireless signals, easily leading to cross-floor signal penetration. These architectural layout characteristics directly affect the stability and accuracy of personnel location signals.

[0003] Current hotel personnel dynamic management systems mostly rely on location and dispatch technologies that infer location based on the strength of wireless signals received by base stations, neglecting the influence of structures such as elevator shafts and corridor waveguides on the signal. These simplistic positioning methods fail to adequately consider the impact of complex hotel building structures on the positioning process, resulting in insufficient positioning accuracy. Consequently, the dispatching process cannot accurately identify candidate employees located on the target floor and in the target corridor area. This often leads to tasks being assigned to personnel not on the corresponding floor or far from the target area, significantly reducing dispatching efficiency, delaying task execution, and impacting customer experience and the overall quality of hotel operations and services. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies, such as insufficient accuracy in determining the location of hotel staff due to interference from building structure, and inaccurate work assignment screening. Therefore, this invention proposes a dynamic staff management method for hotel management.

[0005] To address the problems existing in the prior art, the present invention adopts the following technical solution: A method for dynamic personnel management in hotel management, comprising: S1. Obtain the building information of the hotel, divide the base stations into zones based on the building information of the hotel, and calculate the structured indicators based on the received signal strength of the base stations in the zones. The structured indicators include: shaft energy ratio, vertical dominance ratio and corridor waveguide anisotropy factor. S2. Calculate the relative height value of air pressure based on the current floor air pressure, and form a discrete floor prior distribution based on the relative height value of air pressure; S3. Input the received signal strength, structured index and air pressure relative height value of the base station into the neural network to obtain the posterior distribution of the floor and the posterior distribution of the corridor location unit; S4. Based on the discrete floor prior distribution and floor posterior distribution, a uniform floor posterior distribution is formed, and the floor determination result is generated according to the uniform floor posterior distribution. S5. Determine the structured indicators and the consistent floor posterior distribution, and generate floor consistency labels based on the determination results. S6. Generate work assignment instructions based on the floor determination results, floor consistency labels, and the posterior distribution of corridor location units.

[0006] Preferably, the base stations are partitioned based on the hotel's building information, including: Obtain the hotel's architectural information, which includes: building floor plan and floor information; Based on the hotel's architectural layout and floor information, the base stations are divided into the following sets: base station set for this floor, base station set for adjacent floors above and below, base station set for the elevator shaft neighborhood, base station set along the main axis of the corridor, and base station set along the partition wall.

[0007] Preferably, the structured index is calculated based on the received signal strength of the partitioned base station, including: Obtain the received signal strength of the base station; The shaft energy ratio is obtained by comparing the sum of the received signal strengths of the neighboring base stations in the elevator shaft with the sum of the received signal strengths of all base stations. The vertical dominance ratio is obtained by comparing the maximum received signal strength of the base station sets on adjacent floors with the maximum received signal strength of the base station on the current floor. The difference in signal strength is obtained by comparing the average received signal strength of the base station set along the main axis of the corridor with the average received signal strength of the base station set along the partition wall. The corridor waveguide anisotropy factor value is obtained by the ratio of the mean difference in intensity to the standard deviation of the received signal intensity of all base stations.

[0008] Preferably, the relative height value of air pressure is calculated based on the current floor air pressure, and a discrete floor prior distribution is formed based on the relative height value of air pressure, including: The relative height of air pressure is obtained by logarithmic calculation based on the air pressure value of the current floor and the reference air pressure value at the same location. Divide the relative air pressure height value by the floor height and round down to obtain the discrete floor value; A prior distribution of discrete floors in the form of a normal distribution is constructed based on discrete floor values.

[0009] Preferably, the received signal strength, structured indices, and relative air pressure height values ​​of the base station are input into a neural network to obtain the posterior distribution of the floors and the posterior distribution of the corridor location units, including: The feature vector is composed of the set of received signal strength of all base stations, the proportion of shaft energy, the vertical dominance ratio, the corridor waveguide anisotropy factor, and the relative height of air pressure. Normalize the feature vectors; The normalized feature vectors are input into the shared feature extraction network to obtain intermediate feature representations; Based on intermediate feature representation, the floor posterior distribution and the corridor location unit posterior distribution are generated respectively by setting independent floor classification output heads and corridor location unit output heads; During the training phase, a weighted cross-entropy loss function is used to set sample weights for risk samples based on shaft energy ratio, vertical dominance ratio, and corridor waveguide anisotropy factor.

[0010] Preferably, a uniform floor posterior distribution is formed based on the discrete floor prior distribution and the floor posterior distribution, and a floor determination result is generated based on the uniform floor posterior distribution, including: The floor posterior distribution and the discrete floor prior distribution are multiplicatively fused and normalized to form a uniform floor posterior distribution. The floor number corresponding to the highest probability of the uniform floor posterior distribution is used as the floor determination result.

[0011] Preferably, the structured indicators and the consistent posterior distribution of floors are judged, and floor consistency labels are generated based on the judgment results, including: The maximum probability of the uniform floor posterior distribution is compared with the preset floor confidence threshold. If the maximum probability is not less than the preset floor confidence threshold, the confidence level is determined to be met; otherwise, the confidence level is determined to be unmet. The system determines whether the set conditions are met based on structured indicators. If the conditions are met, it is determined that there is a structured risk; otherwise, it is determined that there is no structured risk. The setting conditions are: the shaft energy ratio is greater than the preset shaft energy ratio threshold and the corridor waveguide anisotropy factor is greater than the preset corridor waveguide factor threshold, or the vertical dominance ratio is greater than the preset vertical dominance threshold. When the confidence level is met and there is no structured risk, a normal floor consistency label is generated.

[0012] Preferably, the dispatch instruction is generated based on the floor determination result, floor consistency labels, and posterior distribution of corridor location units, including: Select candidate employees whose floor consistency label is normal and whose floor determination result is consistent with the target task floor to form a feasible task assignment set; For each employee in the feasible set, based on the posterior distribution of the corridor location unit, the posterior probability of the employee in the target corridor location unit is extracted, and the negative logarithm of the posterior probability is used as the dispatching value. Select the employee with the lowest cost from the feasible set as the final assignment target; Generate a dispatch instruction that includes the dispatch object, the target task floor, and the target corridor location unit.

[0013] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention partitions base stations based on hotel building information and calculates structured indices including shaft energy ratio, vertical dominance ratio, and corridor waveguide anisotropy factor. Simultaneously, it calculates the relative height of air pressure based on the current floor air pressure, forming a discrete floor prior distribution. Then, it inputs the base station received signal strength, structured indices, and relative height of air pressure into a neural network to obtain the floor posterior distribution and the corridor location unit posterior distribution. Subsequently, it generates floor determination results by fusing the prior and posterior distributions, generates floor consistency labels based on the determinations, and finally generates dispatch instructions. This fully covers the entire process from acquiring basic positioning data and fusing multi-source information to generating dispatch instructions, effectively addressing the interference of hotel building structure on positioning signals and solving the core problems of insufficient accuracy in floor and corridor location determination and inaccurate dispatch screening.

[0014] 2. This invention refines base stations into sets such as the current floor, adjacent floors above and below, elevator shaft neighborhood, corridor main axis direction, and partition wall direction. This allows the calculation of structured indicators to accurately correlate the signal propagation characteristics of different building areas in the hotel. For example, the signal ratio of base stations in the elevator shaft neighborhood reflects elevator shaft leakage interference, the vertical dominance ratio reflects cross-floor signal penetration, and the corridor waveguide anisotropy factor quantifies the corridor waveguide effect. This can more accurately capture the influence of building structure on positioning, providing more targeted input features for subsequent neural networks, further improving the accuracy of floor identification and corridor location unit determination, optimizing positioning performance from the data source, and avoiding positioning deviations caused by signal interference.

[0015] 3. This invention enhances the network's ability to fit high-interference scenarios by using a weighted cross-entropy loss function to assign weights to risk samples during the neural network training phase. Simultaneously, it integrates air pressure information and base station signal statistics through multiplicative fusion and normalization of discrete floor prior and posterior distributions to reduce the impact of single data fluctuations. Finally, in the dispatching stage, it selects the optimal dispatching target by screening candidate employees with consistent floor labels and using the negative logarithm of the corridor location posterior probability as the cost. This ensures that the dispatching results meet both floor accuracy requirements and are close to the target task location, effectively solving the problem of low dispatching efficiency and improving the practicality and reliability of dynamic hotel personnel management. Attached Figure Description

[0016] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating a method for dynamic personnel management in hotel management, as provided in an embodiment of the present invention. Detailed Implementation

[0017] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0018] Example: This example provides a method for dynamic personnel management in hotel management. See [link to example]. Figure 1 Specifically, including: S1. Obtain the building information of the hotel, divide the base stations into zones based on the building information of the hotel, and calculate the structured indicators based on the received signal strength of the base stations in the zones. The structured indicators include: shaft energy ratio, vertical dominance ratio and corridor waveguide anisotropy factor. In an embodiment of the present invention, architectural information of a hotel is obtained, and base stations are partitioned based on the architectural information of the hotel. Structured indices are calculated based on the received signal strength of the partitioned base stations. These structured indices include: shaft energy percentage, vertical dominance ratio, and corridor waveguide anisotropy factor, including: Obtain the hotel's architectural information, which includes: building floor plan and floor information; Based on the hotel’s architectural layout and floor information, the base stations are divided into the following sets: base station set for this floor, base station set for adjacent floors above and below, base station set for elevator shaft neighborhood, base station set for corridor main axis direction, and base station set for partition wall direction. Specifically, a base station is a fixed node device deployed in the area to be measured in an indoor positioning system, continuously broadcasting or receiving wireless signals. It provides signal acquisition, location reference, and data upload functions for mobile terminals or those wearing tags. A base station typically includes a wireless radio frequency transmitting and receiving module, a communication network interface, and a connection channel to the positioning server, and is positioned at a known geometric location on a hotel floor. The main function of a base station is to provide identifiable reference data for positioning by exchanging signals with mobile devices or measuring parameters such as received signal strength, latency, and angle, thereby assisting in determining the floor, area, or room where a person or terminal is located. Base stations are grouped into sets such as those on the same floor, adjacent floors above and below, elevator shaft neighborhood, corridor main axis, or adjacent walls, reflecting their deployment location and propagation environment within the building structure.

[0019] Specifically, before dividing the base station clusters according to the hotel's building layout and floor information, standardized deployment of base stations must be completed first. Base stations should be deployed in guest room areas at intervals of 15-20 meters to ensure signal coverage without blind spots. An additional base station should be installed at corridor corners and elevator entrances to compensate for signal attenuation at corners. Base stations in back-of-house corridors should be deployed at intervals of 20-25 meters to balance cost and coverage requirements. The bottom of the base station should be 2.5-3.0 meters above the ground and fixed to the corridor wall or ceiling joists, avoiding obstructions such as air conditioning vents and light fixtures. The installation height deviation of base stations on the same floor should not exceed ±0.1 meters to ensure consistent signal propagation paths.

[0020] Specifically, the "base station set for this floor" refers to the group of wireless base stations deployed and serving the floor to be located; the "base station set for adjacent floors" refers to the group of wireless base stations deployed on adjacent floors above and below the target floor; the "base station set for elevator shaft neighborhood" refers to the group of wireless base stations deployed in or near the elevator shaft or stairwell area of ​​a high-rise building, which may affect the positioning results due to the vertical propagation channel in the structure; the "base station set along the main corridor axis" refers to the group of wireless base stations arranged along the main corridor of the hotel, located on the central axis of the corridor or extending along its direction, which may cause propagation anomalies due to the waveguide effect of the corridor; and the "base station set along the partition wall" refers to the group of wireless base stations arranged on the side opposite to the main corridor axis, passing through the partition wall or located on the vertical side of the corridor, which is used to compare the signal characteristics of base stations along the main corridor axis.

[0021] Specifically, the process begins by obtaining the hotel's complete floor plan, information on each floor's distribution, and the coordinates of all indoor positioning wireless signal base stations. Base stations deployed within the target floor are grouped into a single set for that floor. All base stations on the floor above and below the target floor are grouped into a single set, forming adjacent floor sets. Using the axis of each elevator shaft as the center, base stations located in or adjacent to the elevator shaft are grouped into a single set, as the vertical propagation channel formed by the building structure may have a unique impact on wireless signal attenuation and reflection characteristics. Based on the floor plan, the main axis of each floor's corridor is determined, and base stations positioned along this axis or within a preset distance on either side are grouped into a single set, forming a corridor main axis direction base station set. Base stations deployed on the opposite side of the corridor main axis, passing through partition walls, or located vertically on the corridor's side are grouped into a partition wall direction base station set. This set is used for comparative analysis of signal behavior with the corridor main axis direction base station set.

[0022] Obtain the received signal strength of the base station; The shaft energy ratio is obtained by comparing the sum of the received signal strengths of the neighboring base stations in the elevator shaft with the sum of the received signal strengths of all base stations. Specifically, the received signal strength values ​​of each wireless base station deployed on each floor of the hotel are collected. These received signal strength values ​​reflect the signal propagation status between the base station and the mobile terminal. The total received signal strength of all base stations in the elevator shaft neighborhood base station set is calculated, as well as the total received signal strength of all visible base stations. The former is then compared with the latter to obtain the shaft energy ratio value. This shaft energy ratio value reflects the proportion of wireless signal energy carried by the elevator shaft neighborhood base stations to the entire base station network, thus providing a quantitative basis for judging the impact of vertical propagation or leakage channels on the positioning system.

[0023] The vertical dominance ratio is obtained by comparing the maximum received signal strength of the base station sets on adjacent floors with the maximum received signal strength of the base station on the current floor. The difference in signal strength is obtained by comparing the average received signal strength of the base station set along the main axis of the corridor with the average received signal strength of the base station set along the partition wall. The corridor waveguide anisotropy factor value is obtained by the ratio of the mean difference in intensity to the standard deviation of the received signal intensity of all base stations. Specifically, the received signal strength of all base stations in the base station sets of the upper and lower adjacent floors is extracted first, and the received signal strength with the largest value is selected as the maximum received signal strength of the upper and lower adjacent floors. At the same time, the received signal strength of all base stations in the base station set of the current floor is extracted, and the received signal strength with the largest value is selected as the maximum received signal strength of the current floor. The vertical dominance ratio is obtained by dividing the maximum received signal strength of the upper and lower adjacent floors by the maximum received signal strength of the current floor. This ratio is used to reflect whether the signal strength of the adjacent floors exceeds that of the current floor and to indicate the risk of floor confusion.

[0024] Specifically, the received signal strength of all base stations in the main axis direction of the corridor is statistically analyzed and summed to obtain the total strength in the main axis direction. The average received signal strength in the main axis direction is obtained by dividing the total strength in the main axis direction by the number of base stations in the main axis direction of the corridor. Similarly, the received signal strength of all base stations in the wall-connecting direction is statistically analyzed and summed to obtain the total strength in the wall-connecting direction. The average received signal strength in the wall-connecting direction is obtained by dividing the total strength in the wall-connecting direction by the number of base stations in the wall-connecting direction. The difference between the average received signal strength in the main axis direction and the average received signal strength in the wall-connecting direction is obtained by subtracting the average received signal strength in the main axis direction and the average received signal strength in the wall-connecting direction. Finally, the received signal strength of all visible base stations is statistically analyzed. The signal strength is calculated by averaging all received signal strengths. The difference between the received signal strength of each base station and this average value is squared and summed to obtain the sum of squared deviations. The sum of squared deviations is divided by the number of all visible base stations to obtain the variance. The square root of this variance is then taken to obtain the standard deviation of the received signal strength of all base stations. Finally, the difference between the mean strengths is divided by the standard deviation of the received signal strength of all base stations to obtain the corridor waveguide anisotropy factor. This factor reflects the difference in the magnitude of the signal enhancement along the principal axis caused by the corridor waveguide effect relative to the overall signal strength fluctuation, thus providing a quantitative basis for the positioning system to identify corridor propagation anomalies.

[0025] S2. Calculate the relative height value of air pressure based on the current floor air pressure, and form a discrete floor prior distribution based on the relative height value of air pressure; In an embodiment of the present invention, the relative height value of air pressure is calculated based on the current floor air pressure, and a discrete floor prior distribution is formed based on the relative height value of air pressure, including: The relative height of air pressure is obtained by logarithmic calculation based on the air pressure value of the current floor and the reference air pressure value at the same location. Specifically, two barometric pressure sensors are deployed on each floor: one is located at the elevator entrance, close to the core area of ​​personnel activity, to reflect the barometric pressure of the actual activity area; the other is located in the middle of the corridor, avoiding airflow interference sources such as air conditioning vents and doors; and the reference barometric pressure sensor is deployed in the center of the lobby on the first floor of the hotel, unobstructed, well-ventilated, and away from air conditioning vents and revolving doors, to collect the reference barometric pressure value at the same location. Specifically, the air pressure value of the current floor is obtained. This air pressure value is collected in real time by air pressure sensors deployed inside the hotel, reflecting the absolute atmospheric pressure at the current location. Then, the reference air pressure value at the same location is obtained. This reference air pressure value is usually collected and stably recorded over a long period by a fixed reference sensor located on the lowest floor or a reference floor of the building, and is used as a reference starting point for changes in floor air pressure. Based on the air pressure value of the current floor, the reference air pressure value at the same location, and the floor height, the relative air pressure height value is calculated. The formula for calculating the relative air pressure height value is as follows: In the formula, This is the relative altitude value of air pressure. This is an approximate coefficient for room temperature. This is a constant determined by the standard atmospheric model under normal temperature conditions, used to characterize the proportional relationship between air pressure and altitude. In this embodiment, a fixed value can be taken, for example, around 13,500. This is the air pressure value for the current floor. This is the reference air pressure value for the same location; Specifically, the calculation of relative height based on air pressure is based on the law of air pressure variation with altitude, which originates from the pressure distribution formula under atmospheric hydrostatic equilibrium. Within the troposphere, when air is approximated as an ideal gas and the temperature distribution is approximately constant over short vertical distances, the logarithmic relationship between air pressure and altitude can be used to derive a mathematical model of the ratio of altitude difference to air pressure. As altitude increases, air density decreases, and air pressure decreases exponentially with altitude. Therefore, a logarithmic transformation can convert the nonlinear air pressure-altitude relationship into an approximately linear function, thereby achieving a quantitative estimate of floor height. Air pressure values ​​can be measured using air pressure sensors without additional equipment. Combined with known floor heights and reference air pressure, the relative height value can be calculated. This method offers advantages such as low cost and high real-time performance, making it suitable for vertical positioning and floor identification scenarios within complex buildings. The relative height based on air pressure variation accurately reflects the vertical height difference between the measured location and a reference point.

[0026] Divide the relative air pressure height value by the floor height and round down to obtain the discrete floor value; Construct a discrete floor prior distribution in the form of a normal distribution based on discrete floor values; Specifically, the current relative height of air pressure is first obtained. This value is calculated based on the logarithmic relationship between the current floor air pressure and the reference air pressure. According to the hotel's floor structure design, the standard floor height is extracted. The relative height of air pressure is divided by the floor height to obtain a real value representing the relative floor position. Taking the integer part of this real value yields the discrete floor value, which represents the most likely floor number corresponding to the current air pressure. This discrete floor value is used as the mean of a normal distribution, and a preset standard deviation parameter is set. This standard deviation parameter is pre-configured based on the measurement accuracy of the air pressure sensor and the actual interval characteristics of each floor in the hotel, used to quantify the uncertainty range of the discrete floor value. A continuous normal distribution is constructed based on the mean and standard deviation. Then, the probability density values ​​of this normal distribution at each floor number are discretized to obtain the prior probability corresponding to each floor number. These prior probabilities together constitute the discrete floor prior distribution, where the magnitude of the prior probability for each floor number reflects the initial probability of inferring the terminal's location on that floor based on air pressure information. This prior distribution reflects the statistical uncertainty of barometric altitude estimation and provides a consistent prior reference for subsequent fusion of multi-source positioning information.

[0027] S3. Input the received signal strength, structured index and air pressure relative height value of the base station into the neural network to obtain the posterior distribution of the floor and the posterior distribution of the corridor location unit; In embodiments of the present invention, the received signal strength, structured indices, and relative air pressure height values ​​of the base station are input into a neural network to obtain the posterior distribution of floor levels and the posterior distribution of corridor location units, including: The feature vector is composed of the set of received signal strength of all base stations, the proportion of shaft energy, the vertical dominance ratio, the corridor waveguide anisotropy factor, and the relative height of air pressure. Specifically, the received signal strength data of all visible base stations are first collected and arranged sequentially according to a preset base station numbering order, forming an ordered set of received signal strengths. Then, the previously calculated wellhead energy proportion, vertical dominance ratio, corridor waveguide anisotropy factor, and relative pressure altitude are extracted; these are all individual quantification indicators. Using this ordered set of received signal strengths as the initial part of the feature vector, the four quantification indicators are then added sequentially to the received signal strength set in a fixed order: wellhead energy proportion, vertical dominance ratio, corridor waveguide anisotropy factor, and relative pressure altitude. This forms a multidimensional array containing all input features, with a dimension equal to the sum of the number of all visible base stations and the number of the four quantification indicators, effectively describing the key attributes or features of the data.

[0028] Normalize the feature vectors; Specifically, when normalizing the set of received signal strengths in the feature vector, the average value and fluctuation level of all received signal strengths in the set are first calculated during the model training phase. Then, the average value is subtracted from each received signal strength in the set, and the result is divided by the fluctuation level to obtain the normalized received signal strength. When normalizing the shaft energy ratio, the minimum and maximum values ​​of the ratio in the training data are first determined during the training phase. Then, the minimum value is subtracted from the current shaft energy ratio, and the result is divided by the difference between the maximum and minimum values ​​to obtain the normalized shaft energy ratio feature. When normalizing the vertical dominance ratio, the same method as for the shaft energy ratio is used. Based on the minimum and maximum values ​​in the training data, the difference and division operations are used to map them to a unified interval to obtain the normalized vertical dominance ratio feature. When normalizing the corridor waveguide anisotropy factor value, the scaling is performed using the difference and division operations described above, based on the value range in the training data, to obtain the normalized corridor waveguide anisotropy factor feature. When normalizing the relative pressure height value, the minimum and maximum values ​​of this value in the training data are first determined, and then processed using the same difference and division operation procedure to obtain the normalized relative pressure height feature. Finally, the normalized received signal strength set, shaft energy ratio feature, vertical dominance ratio feature, corridor waveguide anisotropy factor feature, and relative pressure height feature are combined in their original order to form a complete normalized feature vector. After normalization, the dimensions of the obtained feature vector in different dimensions are eliminated, and the numerical scales are basically on the same order of magnitude. This is beneficial for the stable convergence of the subsequent neural network shared feature extraction network and avoids unnecessary amplification of parameter updates by features with large dimensions or strong fluctuations, thereby improving the stability and comparability of the calculation results of the floor posterior distribution and the corridor location unit posterior distribution.

[0029] The normalized feature vectors are input into the shared feature extraction network to obtain intermediate feature representations; Specifically, after normalization, the constructed feature vector is loaded into the feature extraction module as input data for the neural network. This feature extraction module preferably employs a feedforward multi-layer neural network structure, comprising an input layer, several hidden layers, and a convergence layer for outputting intermediate feature representations. Each layer abstracts the input features step-by-step through fully connected operations and nonlinear activation operations. The dimension of the input layer is consistent with the dimension of the normalized feature vector. When the number of visible base stations is N, the number of neurons in the input layer is N plus 4, where 4 corresponds to four structured indicators: wellbore energy ratio, vertical dominance ratio, corridor waveguide anisotropy factor, and relative air pressure altitude. Preferably, the shared feature extraction network includes three hidden layers: the first hidden layer has 512 neurons, the second hidden layer has 512 neurons, and the third hidden layer has 256 neurons. The hidden layers are connected by fully connected weight matrices, and after each hidden layer, affine transformation, batch normalization, and linear rectified activation are performed sequentially, enabling the network to introduce sufficient nonlinear expressive power while maintaining numerical stability. Furthermore, an inactivation operation is set at the output of each hidden layer, randomly zeroing out the outputs of some neurons. The inactivation ratio can be set to, for example, 0.3, to suppress the tendency to overfit to a small number of training samples during training. The input normalized feature vector passes through the above-mentioned layer structure in sequence, undergoing weighting, nonlinear activation, and regularization at each layer. Redundant information is gradually compressed, and key information related to floor and corridor location determination is highlighted. Finally, a fixed-dimensional intermediate feature representation vector is extracted at the convergence layer. The dimension of the intermediate feature representation can be set to, for example, 128 dimensions. This intermediate feature representation is a compact vector expression form containing multi-dimensional abstract information. It simultaneously encodes the base station received signal strength distribution pattern, shaft energy ratio changes, vertical dominance ratio, directional differences indicated by corridor waveguide anisotropy factor values, and floor height information represented by air pressure relative height values ​​in a unified feature space, thus possessing high discriminative ability. By iteratively optimizing the weight parameters of the shared feature extraction network using a large number of labeled samples during the training phase, the network can automatically learn the correlation and importance between different input dimensions and adaptively adjust the contribution weights of various structured indicators. This enables the intermediate feature representation to have good sharing and generalization capabilities across different tasks, providing a unified, stable, and information-rich input foundation for subsequent multi-task classification prediction outputs.

[0030] Based on intermediate feature representation, the floor posterior distribution and the corridor location unit posterior distribution are generated respectively by setting independent floor classification output heads and corridor location unit output heads; Specifically, the intermediate feature representation is simultaneously input to two independently configured output heads. Both output heads employ a fully connected structure with normalized probability transformation, but their output dimensions and the prediction targets they focus on differ. The floor classification output head contains a fully connected layer. The input dimension of this fully connected layer is consistent with the dimension of the intermediate feature representation, and the output dimension is equal to the total number of floors in the hotel. Preferably, it is based on the number of floors that need to be distinguished in the actual business management of the hotel. For example, when the hotel has fifty manageable floors, the output dimension is fifty. The fully connected layer first maps the intermediate feature representation to a feature space corresponding to the number of floors, obtaining the unnormalized score value corresponding to each floor. Then, a softmax activation operation is performed on the score vector, so that the value of each element in the output vector is between zero and one, and the sum of all elements is one. The resulting vector is the posterior distribution of the floors, where each element corresponds to the posterior probability of the terminal being located on the corresponding floor. This distribution represents the probability estimate of the terminal belonging to each floor under the current input feature vector, reflecting the confidence level of the floor determination result. The corridor location unit output head also contains a fully connected layer. Its input dimension is consistent with the dimension of the intermediate feature representation, and its output dimension is equal to the total number of location units in the corridor area. Preferably, each floor corridor is discretized according to the hotel floor plan. For example, each floor corridor is divided into several continuous segments and each segment is assigned a unique number. After the fully connected layer maps the intermediate feature representation to the feature space corresponding to the number of location units, it performs a softmax activation operation on the output vector. The resulting vector is the posterior distribution of the corridor location units, where each element corresponds to the posterior probability of the terminal being located in the corresponding corridor location unit. This distribution represents the probability distribution of the corridor location unit corresponding to the current input feature vector and is used to make fine-grained determinations of the corridor location of employees on the same floor. To ensure that the two prediction tasks can share intermediate features while optimizing independently, the fully connected layers of the two output heads maintain their own weight parameters and bias parameters during training and update them independently according to their respective loss terms during the backpropagation stage. This maintains the constraints of the floor classification task on the intermediate features, as well as the constraints of the corridor location unit classification task on the intermediate features. With this multi-task structure, floor prediction and corridor location unit prediction can complete the high-rise determination separately based on shared underlying features. When the number of floors, corridor division granularity or hotel layout is adjusted, only the output dimension and label mapping relationship need to be adjusted according to the new business requirements. There is no need to make significant changes to the shared feature extraction network structure, which facilitates engineering implementation and later maintenance.

[0031] During the training phase, a weighted cross-entropy loss function is used to set sample weights for risk samples of shaft energy proportion, vertical dominance ratio, and corridor waveguide anisotropy factor. Specifically, during the training phase of the neural network, a training sample set is first constructed based on historical positioning data. The distributions of shaft energy percentage, vertical dominance ratio, and corridor waveguide anisotropy factor values ​​are statistically analyzed in both incorrect and correct positioning samples. Risk thresholds for the shaft energy percentage, vertical dominance ratio, and corridor waveguide anisotropy factor values ​​are then determined based on the quantiles of these distributions. For example, the 80th or 90th percentile values ​​of these indicators in the incorrect samples can be used as the corresponding risk thresholds, ensuring that the thresholds cover most high-risk scenarios. For each sample input into the training set, the shaft energy percentage, vertical dominance ratio, and corridor waveguide anisotropy factor values ​​are compared with the shaft energy risk threshold, the vertical dominance ratio, and the waveguide factor risk threshold, respectively. If the shaft energy percentage, vertical dominance ratio, or corridor waveguide anisotropy factor value is greater than the preset shaft energy risk threshold, vertical dominance ratio, or waveguide factor risk threshold, the sample is determined to be a risk sample. The sample weights are determined based on the number of times a sample simultaneously meets the threshold indicators. Samples with two or more of the three indicators exceeding the corresponding risk threshold are classified as high-risk samples, and their corresponding weights are set to high-weight values. Samples with only one of the three indicators exceeding the corresponding risk threshold are classified as medium-risk samples, and their corresponding weights are set to medium-weight values. Samples with none of the three indicators exceeding the corresponding risk threshold are classified as non-risk samples, and their corresponding weights are set to basic weight values. To avoid excessively large loss values ​​during training that could lead to numerical instability, the high-weight, medium-weight, and basic weight values ​​can be limited to a preset finite range. Furthermore, the sum of all weights is normalized in each training batch to ensure that the proportion of samples with different risk levels in the overall loss remains within a reasonable range. The training loss is calculated using a weighted cross-entropy loss function, which consists of two parts: floor classification loss and corridor location unit classification loss. The floor classification loss is calculated by multiplying the cross-entropy of the predicted posterior distribution of each sample's floor with the actual floor label by the sample's weight. The corridor location unit classification loss is calculated by multiplying the cross-entropy of the predicted posterior distribution of each sample's corridor location with the actual location unit label by the sample's weight. The two losses are summed according to a preset ratio to obtain the total loss. The backpropagation algorithm is then used to update the weight parameters of the shared feature extraction network and the two output heads using the total loss. This allows the network to better fit risk samples during training, gradually reducing the classification error rate in high-risk scenarios such as elevator shaft leaks, vertical interference between floors, and corridor waveguide effects. This significantly improves the positioning accuracy and overall robustness in high-risk scenarios.

[0032] S4. Based on the discrete floor prior distribution and floor posterior distribution, a uniform floor posterior distribution is formed, and the floor determination result is generated according to the uniform floor posterior distribution. In an embodiment of the present invention, a consistent floor posterior distribution is formed based on the discrete floor prior distribution and the floor posterior distribution, and a floor determination result is generated based on the consistent floor posterior distribution, including: The floor posterior distribution and the discrete floor prior distribution are multiplicatively fused and normalized to form a uniform floor posterior distribution. The floor number corresponding to the highest probability of the uniform floor posterior distribution is used as the floor determination result. Specifically, the probability value of each floor in the posterior distribution is multiplied by the probability value in the corresponding discrete prior distribution to obtain the fused probability distribution. This multiplicative fusion operation effectively combines statistical information from base station signal reception with information such as barometric pressure and altitude, improving the accuracy of the final floor identification. The fused probability distribution is then normalized by summing the fused probability values ​​of all floors and dividing the fused probability of each floor by the sum, ensuring that the final probability distribution meets the probability normalization requirement, with a sum of 1. The normalized distribution is the consistent posterior floor distribution, which integrates the influence of posterior and prior information, and can more accurately reflect the probability of each floor. Finally, the floor number corresponding to the maximum probability value in the consistent posterior floor distribution is used to determine the final floor determination result. This floor number is the most likely floor identified by the system and serves as the basis for subsequent task scheduling and personnel positioning.

[0033] S5. Determine the structured indicators and the consistent floor posterior distribution, and generate floor consistency labels based on the determination results. In an embodiment of the present invention, the structured indicators and the consistent floor posterior distribution are determined, and a floor consistency label is generated based on the determination result, including: The maximum probability of the uniform floor posterior distribution is compared with the preset floor confidence threshold. If the maximum probability is not less than the preset floor confidence threshold, the confidence level is determined to be met; otherwise, the confidence level is determined to be unmet. Specifically, the highest probability value is extracted from the uniform floor posterior distribution. This value represents the highest confidence level that the terminal is located on the corresponding floor, indicating the most likely floor inferred based on the current input information. A preset floor confidence threshold is set, determined based on the probability distribution characteristics of correct judgment cases in historical positioning data. This threshold is used to measure the reliability of the floor judgment result. For example, the highest probability value of the uniform floor posterior distribution is extracted from all correct judgment cases to form a probability dataset. The maximum, minimum, mean, median, standard deviation, and 85th, 90th, and 95th percentiles are calculated for this dataset. By plotting frequency distribution histograms and cumulative distribution curves, the central trend of the dataset is observed. Typically, the highest probability value of correct judgment cases will be concentrated in the range of 0.7-0.9. Finally, the threshold is determined based on the hotel's positioning accuracy requirements. If it is required that more than 90% of the correct judgment cases can pass the confidence standard, then the 90th percentile of the probability dataset is selected as the floor confidence threshold. The extracted maximum probability is compared with a preset floor confidence threshold. If the maximum probability is greater than or equal to the preset floor confidence threshold, the reliability of the current floor determination result meets the preset requirements and is judged as meeting the confidence standard. If the maximum probability is less than the preset floor confidence threshold, the reliability of the current floor determination result does not meet the preset requirements and is judged as not meeting the confidence standard. This judgment result will directly affect the subsequent generation of floor consistency labels, ensuring that only when the confidence level is high will the floor result be considered valid and further used for subsequent dispatching or positioning tasks.

[0034] The system determines whether the set conditions are met based on structured indicators. If the conditions are met, it is determined that there is a structured risk; otherwise, it is determined that there is no structured risk. The setting conditions are: the shaft energy ratio is greater than the preset shaft energy ratio threshold and the corridor waveguide anisotropy factor is greater than the preset corridor waveguide factor threshold, or the vertical dominance ratio is greater than the preset vertical dominance threshold. Specifically, based on the acquired structured indicators, including shaft energy ratio, corridor waveguide anisotropy factor, and vertical dominance ratio, these indicators are first compared item by item to determine whether they meet the set conditions. These conditions include two scenarios: the first is that the shaft energy ratio is greater than a preset shaft energy ratio threshold and the corridor waveguide anisotropy factor is greater than a preset corridor waveguide factor threshold; the second is that the vertical dominance ratio is greater than a preset vertical dominance threshold. The shaft energy ratio threshold is determined based on the characteristics of the shaft energy ratio in samples from historical positioning where floor misjudgments were caused by elevator shaft leakage. For example, 500 historical samples of floor misjudgments due to elevator shaft signal leakage are collected. If an employee is actually on the third floor, the positioning result might be the second or fourth floor. The mean and standard deviation of the shaft energy ratio for each sample are calculated, and the mean and standard deviation are used as the shaft energy ratio threshold. The corridor waveguide factor threshold is based on the characteristics of the shaft energy ratio in samples from historical positioning where location misjudgments were caused by corridor waveguide effects. The characteristics of the corridor waveguide anisotropy factor value are determined by, for example, collecting 800 sets of cross-floor signal interference samples. If an employee is located on the fifth floor, and the signal strength received by the base station on the adjacent fourth or sixth floor is greater than that on the fifth floor, the ratio of the maximum received signal strength of the base station on the upper and lower adjacent floors to the maximum received signal strength of the base station on the current floor is calculated, and the median of the ratio is taken as the corridor waveguide factor threshold. The vertical dominance threshold is determined based on the characteristics of the vertical dominance ratio in samples of misjudgment of floors due to cross-floor signal interference in historical positioning. For example, 600 sets of corridor location misjudgment samples are statistically analyzed. If an employee is located in section A of corridor, and the positioning result is the corridor waveguide anisotropy factor value of section B of corridor, the factor values ​​are sorted in descending order, and the top 10% quantile is taken as the vertical dominance threshold to accurately identify strong waveguide interference scenarios. If any condition is met, it is determined that there is a structured risk in the current situation, which means that there are environmental factors that may affect the accuracy of the positioning system, such as the propagation characteristics of elevator shafts or corridors that may lead to floor misjudgment. Otherwise, if none of the above conditions are met, it is determined that there is no structured risk, indicating that the current signal propagation environment is in line with normal expectations and the positioning system risk is low. This determination will serve as the basis for generating floor consistency labels, ensuring that the system conducts reasonable risk assessments based on environmental changes.

[0035] When the confidence level is met and there is no structural risk, a normal floor consistency label is generated; Specifically, the previously determined confidence level and structured risk level are first obtained. The confidence level is either "confidence meets the standard" or "confidence does not meet the standard," and the structured risk level is either "structured risk exists" or "structured risk does not exist." These two levels are then jointly assessed. If the confidence level is "confidence meets the standard" and the structured risk level is "structured risk does not exist," it indicates that the current floor determination result has sufficient reliability and is not significantly affected by structured factors such as elevator shaft leakage, corridor waveguide effects, or cross-floor signal interference. The floor information is determined to be consistent with the actual scenario. At this point, a normal floor consistency label is generated to identify this consistency. This label is used in the subsequent work assignment generation process to screen reliable candidate employee information.

[0036] S6. Generate work assignment instructions based on the floor determination results, floor consistency labels, and the posterior distribution of corridor location units. In an embodiment of the present invention, a dispatch instruction is generated based on the floor determination result, the floor consistency label, and the posterior distribution of the corridor location units, including: Select candidate employees whose floor consistency label is normal and whose floor determination result is consistent with the target task floor to form a feasible task assignment set; Specifically, the target floor for the current task needs to be determined, and the dynamic management information of all available candidate employees within the hotel is obtained. This information includes the floor consistency tag and floor determination result for each candidate employee. The dynamic management information of each candidate employee is checked one by one. First, it is determined whether their floor consistency tag is normal, and then it is determined whether the floor number corresponding to their floor determination result matches the target task floor number. Candidate employees who simultaneously meet the criteria of a normal floor consistency tag and a floor determination result matching the target task floor are selected. The identification information of these selected candidate employees is summarized and organized to form a feasible dispatch set for subsequent work selection.

[0037] For each employee in the feasible set, based on the posterior distribution of the corridor location unit, the posterior probability of the employee in the target corridor location unit is extracted, and the negative logarithm of the posterior probability is used as the dispatching value. Select the employee with the lowest cost from the feasible set as the final assignment target; Generate a dispatch instruction that includes the dispatch object, the target task floor, and the target corridor location unit; Specifically, for each employee in the feasible set, based on the posterior distribution of that employee in the target corridor location unit, the posterior probability of that employee in the target corridor location unit is extracted. The natural logarithm of this posterior probability value is then calculated, and the result is negative. The resulting value is the assignment cost of that employee. A higher posterior probability corresponds to a lower assignment cost, indicating a higher degree of matching with the target location unit and a greater probability of meeting the task requirements. All employees in the feasible set are iterated through, and the assignment cost of each employee is calculated and sorted according to the value. Finally, the employee with the lowest cost is selected as the final assignment recipient, meaning that this employee has the highest degree of matching with the target corridor location unit and is the most suitable person to perform the current task.

[0038] Specifically, the process involves obtaining the identification information of the final identified task assignee, which is a unique employee ID or name. Simultaneously, it extracts the target task floor number (consistent with the hotel's floor numbering system) and the specific division identifier of the target corridor location unit, which is the number or name of the corresponding corridor area division unit. Following a preset instruction format, the task assignee's identification information, the target task floor number, and the target corridor location unit division identifier are sequentially integrated. The task assignee's identification information is placed at the beginning of the instruction to clearly identify the task executor; the target task floor number follows, indicating the floor where the task is located; and the target corridor location unit division identifier is at the end, precisely specifying the specific corridor area within that floor. After integration, a structured electronic task assignment instruction is generated. This instruction can be transmitted to the task assignee through the hotel's internal communication system, ensuring that the task assignee clearly obtains the necessary executor information, task floor information, and specific corridor location information.

[0039] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for dynamic personnel management in hotel management, characterized in that, Includes the following steps: S1. Obtain the building information of the hotel, divide the base stations into zones based on the building information of the hotel, and calculate the structured indicators based on the received signal strength of the base stations in the zones. The structured indicators include: shaft energy ratio, vertical dominance ratio and corridor waveguide anisotropy factor. S2. Calculate the relative height value of air pressure based on the current floor air pressure, and form a discrete floor prior distribution based on the relative height value of air pressure; S3. Input the received signal strength, structured index and air pressure relative height value of the base station into the neural network to obtain the posterior distribution of the floor and the posterior distribution of the corridor location unit; S4. Based on the discrete floor prior distribution and floor posterior distribution, a uniform floor posterior distribution is formed, and the floor determination result is generated according to the uniform floor posterior distribution. S5. Determine the structured indicators and the consistent floor posterior distribution, and generate floor consistency labels based on the determination results. S6. Generate work assignment instructions based on the floor determination results, floor consistency labels, and the posterior distribution of corridor location units.

2. The personnel dynamic management method for hotel management according to claim 1, characterized in that, The base stations are zoned based on the hotel's building information, including: Obtain the hotel's architectural information, which includes: building floor plan and floor information; Based on the hotel's architectural layout and floor information, the base stations are divided into the following sets: base station set for this floor, base station set for adjacent floors above and below, base station set for the elevator shaft neighborhood, base station set along the main axis of the corridor, and base station set along the partition wall.

3. The personnel dynamic management method for hotel management according to claim 1, characterized in that, Structured metrics are calculated based on the received signal strength of the base stations in the partition, including: Obtain the received signal strength of the base station; The shaft energy ratio is obtained by comparing the sum of the received signal strengths of the neighboring base stations in the elevator shaft with the sum of the received signal strengths of all base stations. The vertical dominance ratio is obtained by comparing the maximum received signal strength of the base station sets on adjacent floors with the maximum received signal strength of the base station on the current floor. The difference in signal strength is obtained by comparing the average received signal strength of the base station set along the main axis of the corridor with the average received signal strength of the base station set along the partition wall. The corridor waveguide anisotropy factor value is obtained by the ratio of the mean difference in intensity to the standard deviation of the received signal intensity of all base stations.

4. The personnel dynamic management method for hotel management according to claim 1, characterized in that, Calculate the relative height of air pressure based on the current floor air pressure, and form a discrete floor prior distribution based on the relative height of air pressure, including: The relative height of air pressure is obtained by logarithmic calculation based on the air pressure value of the current floor and the reference air pressure value at the same location. Divide the relative air pressure height value by the floor height and round down to obtain the discrete floor value; A prior distribution of discrete floors in the form of a normal distribution is constructed based on discrete floor values.

5. The personnel dynamic management method for hotel management according to claim 1, characterized in that, The received signal strength, structured parameters, and relative air pressure at the base station are input into a neural network to obtain the posterior distribution of the floor level and the posterior distribution of the corridor location units, including: The feature vector is composed of the set of received signal strength of all base stations, the proportion of shaft energy, the vertical dominance ratio, the corridor waveguide anisotropy factor, and the relative height of air pressure. Normalize the feature vectors; The normalized feature vectors are input into the shared feature extraction network to obtain intermediate feature representations; Based on intermediate feature representation, the floor posterior distribution and the corridor location unit posterior distribution are generated respectively by setting independent floor classification output heads and corridor location unit output heads; During the training phase, a weighted cross-entropy loss function is used to set sample weights for risk samples based on shaft energy ratio, vertical dominance ratio, and corridor waveguide anisotropy factor.

6. The personnel dynamic management method for hotel management according to claim 1, characterized in that, A uniform floor posterior distribution is formed based on the discrete floor prior distribution and the floor posterior distribution, and a floor determination result is generated based on the uniform floor posterior distribution, including: The floor posterior distribution and the discrete floor prior distribution are multiplicatively fused and normalized to form a uniform floor posterior distribution. The floor number corresponding to the highest probability of the uniform floor posterior distribution is used as the floor determination result.

7. The personnel dynamic management method for hotel management according to claim 1, characterized in that, The structured indicators and the consistent posterior distribution of floors are judged, and floor consistency labels are generated based on the judgment results, including: The maximum probability of the uniform floor posterior distribution is compared with the preset floor confidence threshold. If the maximum probability is not less than the preset floor confidence threshold, the confidence level is determined to be met; otherwise, the confidence level is determined to be unmet. The system determines whether the set conditions are met based on structured indicators. If the conditions are met, it is determined that there is a structured risk; otherwise, it is determined that there is no structured risk. The setting conditions are: the shaft energy ratio is greater than the preset shaft energy ratio threshold and the corridor waveguide anisotropy factor is greater than the preset corridor waveguide factor threshold, or the vertical dominance ratio is greater than the preset vertical dominance threshold. When the confidence level is met and there is no structured risk, a normal floor consistency label is generated.

8. The personnel dynamic management method for hotel management according to claim 1, characterized in that, Based on the floor determination results, floor consistency labels, and the posterior distribution of corridor location units, dispatch instructions are generated, including: Select candidate employees whose floor consistency label is normal and whose floor determination result is consistent with the target task floor to form a feasible task assignment set; For each employee in the feasible set, based on the posterior distribution of the corridor location unit, the posterior probability of the employee in the target corridor location unit is extracted, and the negative logarithm of the posterior probability is used as the dispatching value. Select the employee with the lowest cost from the feasible set as the final assignment target; Generate a dispatch instruction that includes the dispatch object, the target task floor, and the target corridor location unit.

Images